{ "last_module_idx": 82, "measurement": { "lm_head.linear": null, "model.layers.0.mlp": [ { "accuracy": 0.9146036725295217, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.926093434032641, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9226845314628199, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.923060981850875, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9830354922696164, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9846627304428502, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9857256710529327, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9906071902889955, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9936510998951761, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9930994400852605, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9947583020517701, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9970879817479535, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9971804532565569, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9982006251811981, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9984137277658048, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9987566984797779, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9992322343352594, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.0.self_attn": [ { "accuracy": 0.8902142424332469, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9075412624760678, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9193279304002461, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9443372738988776, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9503695745217173, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9518840156103435, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9650886905820746, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9661802774981448, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9696553851428785, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9716460830286929, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9764224228106047, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9780935002000708, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9796152240351627, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9812940531655362, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9884948612828004, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9904803100385164, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9906667557201887, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9941206990103972, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9974456676527074, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.1.mlp": [ { "accuracy": 0.9808326410619836, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9813314550801328, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9844632054630079, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9854968582328997, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9903440146069777, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9911089500314311, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9925256645993182, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.99496531133589, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9954382922304305, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9950929907591719, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9957021172893675, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9974987495102381, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9978583186472717, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9986437109151953, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9987256181867499, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9990082422369405, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9996334621682763, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.1.self_attn": [ { "accuracy": 0.9803257145379719, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9809459385118986, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9837873170250341, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9869185552785271, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9877700060606003, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9895298402560385, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9892894374696832, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9911264636014637, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9933320879936218, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.993585211273871, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9943512105628064, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9952199133603197, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9946440693579222, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9957159040005583, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9968430750856274, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9980176012766989, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9971417970955372, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9991132987565116, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9993186767556166, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.10.mlp": [ { "accuracy": 0.9565715507457131, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9579362084991053, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9651870821651659, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9673285390201368, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.977991742523093, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9798771205701327, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9828752702788303, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9884562006122187, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9894577664764304, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9887122151098753, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9901579626296696, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.994212607019826, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9950745921385916, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9968106223172263, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9970387136073489, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9977118733682131, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9991458052475202, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.10.self_attn": [ { "accuracy": 0.9678778773859927, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.968979321028057, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9717913489592703, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9796436774103265, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9837091565132141, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9840335610665774, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.989910677075386, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9903162341368826, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9890433879275071, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9902301611084687, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9916346112364217, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9922925178941927, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9927690319324795, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.993398799708015, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9959352392899362, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9966346052916426, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.997216058796958, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9980914771164718, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9991404917091131, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.11.mlp": [ { "accuracy": 0.9566346532420108, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9579700482519049, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9649937027379086, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9670185697706122, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9780616368118086, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9800019766155043, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9828334544834337, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9885100385076121, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9895417149129667, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9887194241348066, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9902074258578452, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.994207805708835, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.995095882760851, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9968155489156121, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9970171271186126, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9976565528072809, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.999150249910982, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.11.self_attn": [ { "accuracy": 0.9671865011516371, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9682522510227404, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9707271488089311, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9776127307038558, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9822284704760501, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9826375716610959, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9876322260028437, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9881160745495244, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9900295781461816, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9897178191887704, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9915468669251392, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.991921507214245, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9923238573889983, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9927519724557274, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9957240878751403, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9964727037831357, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9968908795792806, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9978229099590528, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9991362777195478, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.12.mlp": [ { "accuracy": 0.9569312710511058, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9582051603417647, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9649120694712588, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9668933655086317, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9781355450027868, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9800489717408231, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9828519899594156, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9885261803865433, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9895351442851519, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9887768908550865, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9902249684459284, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9942347587723481, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9951042050593778, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.996818973241668, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9970529087279972, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9976669112710577, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9991342737094352, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.12.self_attn": [ { "accuracy": 0.9684976057002419, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9693810814305356, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9722787737846375, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9789447878536425, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9829498121612951, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9834213837196952, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9878154538179699, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9884170905539864, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9902055153721258, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9908368642392912, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9917238237042176, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9924568596639132, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9927834719419479, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9932980607998999, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9959445917292645, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9966345174532187, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9969952024127308, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9977936774100128, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9992118314104644, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.13.mlp": [ { "accuracy": 0.9552545547485352, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9566639630418075, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9634791204803869, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.965497120430595, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9773712722878707, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9793075777982411, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.982183857967979, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9881337112502048, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9891702391599354, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9883948313562494, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9898852611842909, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9940209882824045, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.994932809942647, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9967007288023045, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9969378200016523, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9975629729267798, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9991015727190595, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.13.self_attn": [ { "accuracy": 0.9637941749472367, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9648083228813975, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9677952120178624, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9744153697239725, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9808228784485867, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9812018118406597, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.986068805581645, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9864057694610796, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9888477584249095, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9898571034795359, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9909018474189859, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9913577868750221, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9914683472169074, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9927780047843331, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9951766392118052, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9961675323153797, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9963177301381764, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9975939885173973, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9990940727293491, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.14.mlp": [ { "accuracy": 0.9536153329046149, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9550843395684895, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9622548598992198, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9644144961708471, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9764667357269087, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9784872673059765, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9814996954641844, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9875849093261518, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9886357658787778, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.987860475715838, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9894393047219828, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9937730364893612, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9947063754263678, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9965293113338319, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9968045332321995, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9974551989059699, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9990439772801963, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.14.self_attn": [ { "accuracy": 0.9624935765015452, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9630165821627567, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9652109146118164, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9706897704224837, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.981150558120326, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9806588260751021, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.985657839398635, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9846777351279008, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9850424418323919, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9863854756480769, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9908363662267986, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9908118146030527, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9912829767716559, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9914964766878831, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9951866382831022, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9960481332320916, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9961413497987547, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9977787097817973, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9989197975710818, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.15.mlp": [ { "accuracy": 0.9522772274519268, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9538210755900333, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9614703278792531, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9638475587493495, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9757371293871027, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9778157190272683, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9810886336000342, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9871332739528856, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9882602840662003, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9875600918343193, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9891508394166043, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.993624731506172, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9945690710293619, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9964487191877867, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9967465265408942, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9974604129399124, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9990482456786068, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.15.self_attn": [ { "accuracy": 0.9564402134794938, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.95717832289244, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9601790152098003, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9658156319668418, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9808165776102167, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9810132180389605, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9860316797306663, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9862324360169863, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9862103917096791, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9867930286809018, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9898000665401158, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9905018226096505, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.990569396238578, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9914349380292391, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9945292602244177, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9953243626575721, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.995510673444522, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9972045545123125, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9988978111038083, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.16.mlp": [ { "accuracy": 0.9494576297308269, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9511257127711648, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9593731980574758, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9619873291567752, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9742412567138672, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9764359028715837, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.979992505751158, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9862711257056186, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9874864385316247, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9867882759947526, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9884732122484007, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9932291735159723, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9942312656264556, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9962107370558538, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9965470672438019, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9973325721527401, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9989763000293782, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.16.self_attn": [ { "accuracy": 0.9559140393608495, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9570866164408232, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9581218989271867, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.963792549936395, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9795843566718855, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9798285772925929, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9858072447149377, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9860285379384693, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.986721761916813, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9871267001879843, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9893678230674643, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9880651761042444, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9885315330404985, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9892390550751435, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9945546321963009, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9956295709860953, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9956544912175128, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9973158773623014, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9988779086423548, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.17.mlp": [ { "accuracy": 0.9451589835317511, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9470729451430471, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9560409313754031, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9589567968719884, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9721072316169739, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9745389521121979, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9783763163968137, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9850065676789534, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.986382865592053, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9857141563766881, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9875618423286238, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9926838223871431, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9937740525132731, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9958668935455774, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9962818908848261, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9971377477049828, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9988851202161688, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.17.self_attn": [ { "accuracy": 0.9557533326901888, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9574560240695351, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9611788454808687, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9698379416214792, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.977645257585927, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9779580564875352, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9851778698594946, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9854776467147627, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.985942321388345, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9868116943459762, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9883516375955782, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9892636584608179, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9897905666577188, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9908500078477358, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9943342342188484, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9952974629245306, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9957760097949129, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9972588951258283, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9988742664848503, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.18.mlp": [ { "accuracy": 0.9433004730626157, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9452723264694214, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9546518294434798, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9577015261901052, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9712338855392054, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.973720919144781, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9777472176049885, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9845929020329526, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9859833529120997, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9852786001406217, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9871776637278105, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9924658364371249, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9935826721944307, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9957475901434296, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9961692628107572, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9970488328682748, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9988276190859707, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.18.self_attn": [ { "accuracy": 0.9559288966028314, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9578706935832375, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9617280646374351, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9704369149710003, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9767923590384031, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.977366201187435, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9833401645484724, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9840482206721055, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9868202931002567, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9875387063151911, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9882459358165139, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.98956551049885, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9894162455671712, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9910972745794999, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9940054483319584, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9955604762623185, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9952368885278702, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.997437832779006, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.998660164542104, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.19.mlp": [ { "accuracy": 0.9428346909974751, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9447833299636841, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9540879726409912, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.957111982922805, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9710174202919006, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9735181959051835, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9775311774329135, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9845260538552937, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.985895972502859, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9851573376279128, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9870983848446294, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9924065839303168, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9935510060504863, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9957101054881748, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9961443249332277, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9969912359589025, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9988322365832957, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.19.self_attn": [ { "accuracy": 0.9542441399473893, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9554875963612607, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9595493768390856, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9671793799651296, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.975263365005192, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9771396988316586, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9809197752099288, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9832866772225028, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9866144076773995, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.987097608415704, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9880621848922027, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9897930912281337, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9891719465192995, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9907999744540766, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9934846986281244, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9952719274320101, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9945642579542963, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.997270013352758, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9985763410988607, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.2.mlp": [ { "accuracy": 0.9618985621552718, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9614388472155521, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9593016066049275, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9596736054671438, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9784508033802635, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9884568719487441, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9892690981689253, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9826220336713289, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9902828531829935, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9874366710060521, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.993007201113199, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9955861537080062, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9958217065585288, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.99593664548899, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.998423767717261, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9986168913739292, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9988392429720414, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.2.self_attn": [ { "accuracy": 0.9672725953553852, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9683106823971397, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9748860754464802, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9780709476847398, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9811634233123377, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9841658880836085, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.982715346311268, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9861032068729401, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9900108513079191, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9903992367418188, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9917273772390265, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.993816944721498, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9922312058900532, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9942626466876582, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.995128780211273, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9970098449603507, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9953941196987504, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9987751442546907, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9988421492866779, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.20.mlp": [ { "accuracy": 0.9435623256783736, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9453683401408949, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.954577270307039, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9575854320275157, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9714670244016146, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9738306967835677, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9778377460805994, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.984831806860472, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9861813736589331, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9854520135804227, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9872908349099913, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9925716487984908, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9936461813355747, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.995852880179882, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9962222325174432, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9970505004258532, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9988640684046244, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.20.self_attn": [ { "accuracy": 0.9608393907546997, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9619483383078324, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9663774967193604, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9736068687940899, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.978788015089537, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9798232442454288, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9836536959597939, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9848728572067461, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9878081820513073, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9885236315037075, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.989837048869384, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.991032915680032, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9906421946851831, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9918595748512369, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9944599655113722, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9958739139531788, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9954217586078142, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9976351447403431, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9987669748891341, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.21.mlp": [ { "accuracy": 0.94247389467139, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9442662188881322, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9535060117119237, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9565237948769018, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9708639195090846, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9732448929234555, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9772853380755374, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9845615656752336, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9859393584100824, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9851392256586176, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9870143171988035, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9924237053645285, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.993517807047618, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9957976533394111, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9961516472854113, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9969893678238517, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9988517612218857, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.21.self_attn": [ { "accuracy": 0.9642854050586098, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9653718785235756, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.968754956596776, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9758266304668627, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9813180408979717, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9818955393213975, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9865019007732994, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9871675207426673, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9890647334487814, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9902386579074358, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9907748738401815, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.991699597553203, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9914333592904242, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.992898724581066, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9951651076737204, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.996348279087167, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9961785348622423, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9977487758978417, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9989666623112402, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.22.mlp": [ { "accuracy": 0.9394302870097914, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9413441105892784, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9509604824216742, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9541104530033312, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9692219401660719, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9718218979082609, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9760446642574511, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9836283106552927, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9851093433405224, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9842315488740018, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9862906901459945, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9919252105449375, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9931416856615167, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.995472199822727, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9958707435350669, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9967350189230944, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.998741736145396, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.22.self_attn": [ { "accuracy": 0.9613742138210096, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.962896422335976, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9669490738918907, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.973765630471079, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9795761720130318, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9799875789567044, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9842538755191, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9847013071963662, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9874638444498965, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9879561019571204, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9898600797904166, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9910720335809808, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9907558764282026, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9921262993624336, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9944405281230023, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9959816383688074, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9952289760112762, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9977231757029107, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9987550286674186, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.23.mlp": [ { "accuracy": 0.9370449655934384, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9389941817835757, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9488463872357419, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9520800050936247, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9679587483406067, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9706666155865318, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9750195992620367, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9829147960010328, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.984461596137599, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9836233242561943, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9857338823770222, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9916112415100399, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9928603940888455, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9952836040603487, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9957292785770014, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9966281879889337, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9986932183566847, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.23.self_attn": [ { "accuracy": 0.9665313614042181, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9678907802230433, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9709079045998422, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9770663113970506, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9814375764445255, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9832873814984372, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9857716183913382, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9881167380433333, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9897739965664712, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9901261368864461, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9913568073197415, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9921477068411676, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9921842562524896, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9929926818922946, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9952705875039101, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9964449503704121, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9961631831369901, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9981026406350889, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9989331023473489, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.24.mlp": [ { "accuracy": 0.9374911785125732, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9393593072891235, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9489380936873586, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9520200961514523, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9682941091688055, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9708388981066252, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9750879752008539, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9832308872749931, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9847252337556136, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9838446457135049, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.985851736445176, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9917803006736856, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9929429198566236, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.995451105268378, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.995825793398054, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.996683941859948, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9987597090045088, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.24.self_attn": [ { "accuracy": 0.9619279974385312, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9634955054835269, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9662432545109799, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9719956328994349, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9791069736606196, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9812507033348083, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9833630041072243, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9861891473594465, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9880018791085795, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9886627863896521, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9902322833475313, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9914128458813617, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9916550368070602, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9919691744603609, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9950857225217318, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9960754011806688, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9959626864445837, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9978010299566545, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9989419247170812, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.25.mlp": [ { "accuracy": 0.9383025420339484, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9401164870513112, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9490489583266408, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9518984901277643, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9687281345066271, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9713178433870014, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9753268110124689, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9834712806500887, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9850003985982192, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9840585903117531, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9860542087178481, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9918804694163171, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.993044207754888, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9955072653921027, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9958720909137475, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9966705045417735, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9987521651936205, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.25.self_attn": [ { "accuracy": 0.9647987171223289, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9680633356696681, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9714141927267376, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9775517441724476, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9815655538910314, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9837308008419839, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9858512972530565, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9882415407582333, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9896397684749804, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9904878884553909, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9919308094601882, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9926310338472065, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9927888496925956, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.993489229365399, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9956958133139109, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9966600560828259, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9964553227550105, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9982261514585269, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9990190394026669, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.26.mlp": [ { "accuracy": 0.9388229533245689, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9405575614226491, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9492792267548411, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.952055877760837, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9688584145746733, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9713733259000277, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9752491759626489, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.98364845859377, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.985104153030797, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.984143498696779, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.986110635493931, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9919280825476897, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9930779667277085, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9955700812371153, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9958978616877606, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9966643695768557, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9987950395596655, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.26.self_attn": [ { "accuracy": 0.9718871681313765, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9732925421313235, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9760848111227939, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9808485335425327, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9840332724546131, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9855895905118239, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9877563730666512, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9897383051483255, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9906937291747645, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9913541580501356, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9925815431695235, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9931665141331522, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9933503152508485, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9940078317334777, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9960085714333936, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9970559144490644, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9966776774902093, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9984863057340446, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9990665073457518, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.27.mlp": [ { "accuracy": 0.9378668258064672, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9396259408248098, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9481801923952604, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.950898267720875, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.968343806894202, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9709175950602481, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9747232684963628, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9834101215789193, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9848781055525729, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9838801402794687, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.98589510352988, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.991794121108557, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9929633062136801, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9954986462467595, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9958208485653526, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9965668999050793, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9987657870116987, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.27.self_attn": [ { "accuracy": 0.9720996869237799, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9741642004565189, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9777222150250485, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9815778967581297, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.984770445447219, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9859888883013475, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9880799227639249, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9896134232219896, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9914727556078058, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9916828651177255, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9928733533934543, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9935551756306699, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.993726788774917, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9944021168508028, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9963046167241899, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9972309646637816, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9968760209648233, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9985913369608553, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9991332553327084, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.28.mlp": [ { "accuracy": 0.9361885095897474, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9379724452370092, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9465875311901695, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.949328507247724, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9674614103216874, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9700896739959717, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9739310443401337, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9829001003190091, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9844470557413603, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9834453855690203, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9854970762604162, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9915779687856373, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9927660673856735, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9953738882353431, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9957212672421807, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9964777666487192, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9987332859125576, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.28.self_attn": [ { "accuracy": 0.9661478180634349, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9680281940259432, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9717060296159041, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9774265634386163, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.982243719853853, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9831194265892631, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.98635775164554, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9874325418158582, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9886904202009502, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9898544910706972, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9913365048797507, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9921374775861439, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9919011835989199, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9930000901222229, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9953183629795125, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9964407388316957, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9961266156874204, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9980419538915157, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9988968205687246, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.29.mlp": [ { "accuracy": 0.93642468828904, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9381574078610069, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9465309444226717, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9492019069822211, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9675370548900805, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9701958988842211, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9739087302433816, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9830061385506078, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9845405785660994, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9834866131606855, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9855505071188274, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9915965340639415, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.992797065722315, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9954009499204787, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9957124504603838, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9964354681341272, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.998736264772321, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.29.self_attn": [ { "accuracy": 0.9676822580789265, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9706133321711892, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9738945976683968, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9786307043150851, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9828416360052008, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9844427296989843, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9865858272502297, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9885413231034028, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9903273151109093, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.990753845948922, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9919175991886541, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9928857268471467, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9930532119776073, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9938518452016931, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9956653314201456, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.996839805261085, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9962609688702383, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9981470255083159, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9990488351567796, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.3.mlp": [ { "accuracy": 0.9807592646071785, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9813143140391299, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.984491594527897, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9854990432136937, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9902568341870057, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.991042255571014, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9924485659913013, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9949098809769279, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9953776605819401, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9950311964279727, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9956557460521397, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9974666329983034, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9978330160834288, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9986228875227665, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9987082446092054, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9990026909661921, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9996347319040644, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.3.self_attn": [ { "accuracy": 0.9887831328730834, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9895607907521097, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9912725875252172, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9931462571809166, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9942719575605894, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.994694961529029, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9954774422865165, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9959801521740461, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.996314788335248, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9965109997674039, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9971434626924364, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9974829648670397, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9974471709053767, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9977499111310432, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9985629299557522, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9987837192847541, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9988010064943841, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9994412829217157, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9996901072659775, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.30.mlp": [ { "accuracy": 0.9352090546959325, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9370529212449726, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9452855963456004, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9479547425320274, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9669776998068157, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9696682252381977, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9733174756953591, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9826484118637285, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9841878210243425, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9831892973498294, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9852916912028664, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.991456853715997, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9926685063462508, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9953049546793887, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9956505012355352, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9963645660563519, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9986952136417753, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.30.self_attn": [ { "accuracy": 0.9659312932114852, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9732574726405897, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9776333852818138, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9809478850741136, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9855105265190727, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9855275530564157, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9884415096358249, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9886953344470576, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9903823090227026, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9911009261482641, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9929367852838415, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9933310598134995, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9935287648909971, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9940674399074755, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9962409668062863, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9971673565083429, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9967050724907925, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9985459157706875, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9991469514605246, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.31.mlp": [ { "accuracy": 0.9341773296657362, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9360576742573788, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.944356033676549, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9471418543865806, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9663857096119931, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9691131585522702, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9728449959503976, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9822854776131479, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9838720183623465, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9828900067429793, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9850147617490668, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9912893756439811, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9925252465825332, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.995196387956017, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9955651787550825, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9963051386569676, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9986697859866055, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.31.self_attn": [ { "accuracy": 0.9673204359255339, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9686267438687777, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9720036073734886, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9774690348851053, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9824775065246382, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9837338077394586, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9867079195223356, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9882496415000213, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9897402007328836, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9901459491566608, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9916224934552845, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9924087108750093, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9925031544346559, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9933003571472669, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9956447191928562, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9966266167006994, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9964465298934987, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9980971530864113, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9990746888675188, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.32.mlp": [ { "accuracy": 0.9335096886283473, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9354397874129445, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.943655296375877, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.946466747083162, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9660387133297167, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9688058181812889, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9725331162151537, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.982020649470781, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.983647639814176, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9827006898428264, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9848617613315582, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9912028861673254, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9924511329123848, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9951314224224341, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9955234978544084, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9962653086373681, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9986383379682114, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.32.self_attn": [ { "accuracy": 0.9719625962407965, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9733570437682303, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9771333242717543, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.980706302743209, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9844516296135751, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.985850309070788, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9879308954665535, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.989612349554112, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9910880060572373, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9913769756492815, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9925579075750551, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9934133353986239, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9935140029380196, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9943632883460898, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.995978970276682, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9971577909431959, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.996485388200534, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9984668725611348, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9989494516661293, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.33.mlp": [ { "accuracy": 0.9306512004450748, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9327216838535509, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9414711312243813, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9444206388373124, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9646330224840265, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9675280639999791, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9714524055782118, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9812750377153096, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9829560267297845, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9819941301094858, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9842470178478643, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9908479177638104, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9921427682826394, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9949251511379292, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9953438282797211, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9961257482829847, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9985874090343714, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.33.self_attn": [ { "accuracy": 0.9663940323026556, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.967502844961066, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9724092765858299, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9778510049769753, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9821917477406954, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9829479189295518, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.987065641503585, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.988041069946791, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9893831940073716, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9899332084153828, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9914527237415314, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9920336235510675, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9923944622278214, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.993068299795452, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9957968078945812, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9966086019810877, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9966427770100141, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.998163500310559, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9990967380765238, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.34.mlp": [ { "accuracy": 0.9309952949222765, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.933068639353702, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9417006091067666, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9446627842752557, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9647692849761561, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9676702524486341, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9715544110850284, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9813216250193747, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.982997715473175, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9820375458190316, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9842890877472726, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9908521402823297, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9921665481830898, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9949287301615665, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9953413644903585, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.996122479046646, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9985888136648818, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.34.self_attn": [ { "accuracy": 0.9550836776432238, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.958030568925958, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9624329240698564, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.968410379008243, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9779520238700666, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.97918900847435, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9828225891841086, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9844485050753543, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9868840457577455, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9869980317981619, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.988448147711001, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9903840265775982, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9899111986160278, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9914171962361586, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9938866735288971, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.995780537002965, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9947693108728057, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9975424750070823, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9986745144583677, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.35.mlp": [ { "accuracy": 0.9279767149373105, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9302951160230135, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9390550161662855, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.942192554473877, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9632417747848913, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9662742583375228, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9702680644236112, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9803514308051059, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9821373616394243, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9812754251455006, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9836239689274838, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9904668613484031, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9918264378058282, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9946602653515967, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.99514906892651, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9959642345968046, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9984969322226549, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.35.self_attn": [ { "accuracy": 0.964866042137146, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9682286883655348, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9723897482219496, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9773834300668616, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.982017518658387, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.983099932733335, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9862445106631831, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9874100591007032, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9887479433887884, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9892165284407767, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9913568167310012, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9921488848171736, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9922498684180411, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9932102545311576, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9951920238764662, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9964972208989294, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9958775494443742, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9978924934801302, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.998929121776631, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.36.mlp": [ { "accuracy": 0.9232648924777382, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9256026431133872, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9351558246110615, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9386178004114252, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9606387897541648, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9639326208516171, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9683486536929482, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9789203609290876, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9808507545998222, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.979873200780467, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9824171034913314, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9897153252049496, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9912122762516925, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9942555058943597, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.994752981945088, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9956894422832289, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9983881697254745, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.36.self_attn": [ { "accuracy": 0.9557132752318132, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9565050068654513, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.962658377070176, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9693853353199205, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9781936062009711, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9787191105516333, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9838946932240537, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9844804296368047, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.985106381930803, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9850002793889296, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9886578462625805, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9902314896646299, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9899273569646635, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9915204275595514, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9940714126354769, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9951879978179932, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9950190311984012, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9969529697769567, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9986535586220654, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.37.mlp": [ { "accuracy": 0.9154049722771895, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9185097343043277, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9289309853001645, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9325643401396901, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9564703075509322, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9604757961473966, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9652861168510035, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9767286918665233, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9787665119296626, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9776854326850489, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.980651656263753, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9885206991120389, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9903011635730141, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.993491959414984, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9941225659690405, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.99512146530967, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9981640491046404, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.37.self_attn": [ { "accuracy": 0.9498044189653898, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9516809582710266, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.958022559943952, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9682356777944063, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9740963531167883, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9745617603000841, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9832422184316736, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9836871608307487, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9855864926388389, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9862627104709023, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9871603195604525, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9880982072729814, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9891135261247033, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9900825282460765, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9938911081928956, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9948738593804208, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9955475714645887, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9970816701258484, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9987492339783594, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.38.mlp": [ { "accuracy": 0.903095891601161, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9073334681360345, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9173680920349925, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9210327110792461, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9499946456206472, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9554639364543714, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9600849496690851, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.972636668305648, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9750199772809681, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9741810183776053, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9777927743761163, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9864441956344404, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9886971225864009, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9917229139491132, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9931686300980417, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9941692379744429, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9976976482491744, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.38.self_attn": [ { "accuracy": 0.94539727662739, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9557693067349886, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.960969466912119, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9676955812855771, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9755731726947584, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9763093700534419, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9825342877915031, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9832354407561453, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9853370456319106, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9858393747555582, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9880265656270479, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9890950980939364, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9894867053157405, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9903320332891062, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9937041993988188, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9950928013575705, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9948275689231721, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.997283258916516, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.998585331988962, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.39.mlp": [ { "accuracy": 0.8804321790996351, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.8856545498496607, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.8959678850675884, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9008125066757202, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9387843797081395, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9445681634702181, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9491331514559294, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9661927285947298, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9693506109087091, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9685409257286474, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9728050796609176, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.983763043817721, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9862340531851116, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9903186394980079, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9916336410924008, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9928481649411353, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.996891430334041, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.39.self_attn": [ { "accuracy": 0.9535439390885203, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9556328371951455, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9611960461265162, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.970795612586172, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9785723968556053, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9790178882448297, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9855449701610365, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9860322647973111, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9875878903426623, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9870416741622122, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9893426173611691, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9898676119352642, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9906024626995388, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9913182752697092, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9947315132931659, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9953810298129132, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9959881988010908, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9973575517927346, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9986968495343861, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.4.mlp": [ { "accuracy": 0.9755456039780065, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9762898918829466, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9804606751391762, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9817438031497755, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9876392962116944, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9886461535566732, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9904521099830929, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9935421018224013, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9941252136701032, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.993700523125498, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9944902611406226, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9967886017341363, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9972524601770075, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9982514326509676, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9983600166283155, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9987396498847949, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9995343245958027, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.4.self_attn": [ { "accuracy": 0.9841338443128687, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.98472272408636, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9876134513240111, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9904595088017615, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9914594182842656, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9919744753523877, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.993182012909337, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9937822375642625, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.994538785203507, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9945863897078916, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9958026013091991, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9961148680824983, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9961050521386298, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9964524134993553, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9977376153202433, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.998345598774521, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9980424462180388, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9992005865726816, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9994804143513504, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.5.mlp": [ { "accuracy": 0.9700671277548137, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9709770616732145, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9760693892052299, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9776257056938974, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9848520065608778, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9860844125873164, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9882879625809821, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9920641638730702, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9927906048925299, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9922860123609242, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9932525691233183, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9960707186868316, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9966358005216247, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9978560640623695, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9979957797025379, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9984597981368241, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9994314732030034, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.5.self_attn": [ { "accuracy": 0.9846545740177757, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9852262719681388, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9872581127442812, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9903780308208967, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9919226334283227, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9921528830340034, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9939291010561743, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9941700987125698, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9951227021060491, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9954372209153677, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9959556350582525, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9964209826369035, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.996420894014208, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9968367503269723, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9979888373299649, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9984286470632804, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.998391732965645, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9991392879501769, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9995678352976316, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.6.mlp": [ { "accuracy": 0.9645047940705952, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9656393716209813, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9717388372672232, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9735803729609439, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9821293636372215, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9835792325044933, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9861744406976198, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9906421382176248, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9914799716911817, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9908924220423949, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9920326792880109, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9953616610483119, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9960308572963664, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9974586091151363, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9976357390221796, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9981787528254484, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9993284247619542, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.6.self_attn": [ { "accuracy": 0.9798698472349268, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9806957542896271, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.982775866985321, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9873434493416234, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9896603885449862, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9900021537354118, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9927409557919753, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9931755128659701, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9936676766527327, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9939827428836572, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9948151107681426, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9951178713848716, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9953631719476298, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9956139330016939, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9974096935046347, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.997836991360313, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9980323265649771, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9987858836784175, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9994791876034517, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.7.mlp": [ { "accuracy": 0.9611692961893583, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9623789912775943, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.968944185658505, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9708966048140275, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9804543460670271, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9820405464423331, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9848058333522395, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9898094056468261, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9906901254465705, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9900293734512831, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9912760249878231, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9949145183751458, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9956493150246771, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9972278448311906, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9974011998427542, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9979887869405119, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9992648460167018, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.7.self_attn": [ { "accuracy": 0.9758893565127724, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9768864528128975, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9792004980539021, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9848376669381794, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9874854566235292, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9878244854901966, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9916345233979978, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9920117172755694, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9928674462594484, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.992972698650862, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9937880941127476, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9943162202835083, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9945946939681706, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9951028090558554, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.99702786085637, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.997521376923511, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9978562571892613, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9986202422725526, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9994195588423234, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.8.mlp": [ { "accuracy": 0.9584409374939769, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9596955933068928, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9666724957917866, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9687803136674982, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9789637735015467, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9806931677617525, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9836569757837998, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9890010411802092, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9899656733399943, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9892532809784538, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.990604610035294, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9945166479600104, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9953101155789275, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9970023584993262, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9971988767777619, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9978358449512407, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9992052376466362, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.8.self_attn": [ { "accuracy": 0.9731117832033258, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9745758834638094, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.977521665786442, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9835119780741239, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9860691898747495, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9864617539079565, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9904349312970513, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9908431478236851, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9919336775415822, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9923853003665021, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9932908455007955, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9938535372677603, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9941174874180242, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.994398260979276, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9966611277900244, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9972655196723185, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9975839527814012, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9984499205295977, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9993365132494977, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.9.mlp": [ { "accuracy": 0.9582830385157937, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 492374592, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9595687420744645, "down_proj": { "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "group_size": { "2": 64, "3": 64, "6": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 510724672, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9664769235410189, "down_proj": { "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "3": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 569864704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9685635660824022, "down_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 639496704, "up_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9785271851639998, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 128, "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 721045344, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9804815963694924, "down_proj": { "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "3": 32, "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 740855808, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9834024325797432, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 }, "total_bits": 796587104, "up_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "3": 32, "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9887844143729461, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 910810592, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9896051593517002, "down_proj": { "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "4": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 924225536, "up_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9889225677440041, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 128, "5": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 937477984, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9905055973090624, "down_proj": { "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "4": 32, "5": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 957288448, "up_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.994462293229605, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 128, "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 1153910624, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9952420313891611, "down_proj": { "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "group_size": { "5": 32, "6": 32, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 1173721088, "up_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9969492982092657, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "6": 128, "8": 32 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 1336788832, "up_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9971616064247332, "down_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1380525056, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.997799213975668, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 }, "total_bits": 1505043456, "up_proj": { "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "6": 128, "8": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9991558799029965, "down_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "gate_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 1769284608, "up_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.layers.9.self_attn": [ { "accuracy": 0.9680912243692499, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 111655168, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9692194367709913, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 114997504, "v_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 } }, { "accuracy": 0.971783110969945, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 119288192, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9787669072025701, "k_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "2": 64, "3": 64 }, "scale_bits": 4 }, "total_bits": 139930496, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9833194315433502, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165321856, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9837150385505274, "k_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 128, "4": 128 }, "scale_bits": 4 }, "total_bits": 165487616, "v_proj": { "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "3": 64, "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.988787074622355, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 211983488, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 } }, { "accuracy": 0.989359123142142, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 128 }, "scale_bits": 4 }, "total_bits": 212149248, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9900832787940377, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 64 }, "scale_bits": 4 }, "total_bits": 213960704, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9907121321088389, "k_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 }, "total_bits": 216920576, "v_proj": { "bits": [ 4 ], "bits_prop": [ 1 ], "group_size": { "4": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9913882495541322, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 128, "5": 128 }, "scale_bits": 4 }, "total_bits": 217916416, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9922623799035424, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 219400192, "v_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9924699789599368, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 64, "5": 64 }, "scale_bits": 4 }, "total_bits": 223787264, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 64 }, "scale_bits": 4 } }, { "accuracy": 0.9932598902990943, "k_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "4": 32, "5": 32 }, "scale_bits": 4 }, "total_bits": 226914816, "v_proj": { "bits": [ 5 ], "bits_prop": [ 1 ], "group_size": { "5": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9957555787343728, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 128, "6": 128 }, "scale_bits": 4 }, "total_bits": 274898048, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9965948427194044, "k_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "group_size": { "5": 32, "6": 32 }, "scale_bits": 4 }, "total_bits": 279343616, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9968271004526239, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 }, "total_bits": 316841088, "v_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 128 }, "scale_bits": 4 } }, { "accuracy": 0.9980352866022211, "k_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "o_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "q_proj": { "bits": [ 6 ], "bits_prop": [ 1 ], "group_size": { "6": 32 }, "scale_bits": 4 }, "total_bits": 332263936, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 32 }, "scale_bits": 4 } }, { "accuracy": 0.9990741423282185, "k_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "o_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "q_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 }, "total_bits": 421698688, "v_proj": { "bits": [ 8 ], "bits_prop": [ 1 ], "group_size": { "8": 128 }, "scale_bits": 4 } } ], "model.norm.norm": null } }