{ "bits": 4, "group_size": 128, "sym": true, "data_type": "int", "batch_size": 4, "nsamples": 512, "low_gpu_mem_usage": true, "autoround_version": "0.6.1.dev", "quant_method": "auto-round", "packing_format": "auto_round:auto_gptq", "extra_config": { "model.layers.0.self_attn.q_proj": { "bits": 16 }, "model.layers.0.self_attn.k_proj": { "bits": 16 }, "model.layers.0.self_attn.v_proj": { "bits": 16 }, "model.layers.0.self_attn.o_proj": { "bits": 16 }, "model.layers.0.mlp.router.router": { "bits": 16 }, "model.layers.1.self_attn.q_proj": { "bits": 16 }, "model.layers.1.self_attn.k_proj": { "bits": 16 }, "model.layers.1.self_attn.v_proj": { "bits": 16 }, "model.layers.1.self_attn.o_proj": { "bits": 16 }, "model.layers.1.mlp.router.router": { "bits": 16 }, "model.layers.2.self_attn.q_proj": { "bits": 16 }, "model.layers.2.self_attn.k_proj": { "bits": 16 }, "model.layers.2.self_attn.v_proj": { "bits": 16 }, "model.layers.2.self_attn.o_proj": { "bits": 16 }, "model.layers.2.mlp.router.router": { "bits": 16 }, "model.layers.3.self_attn.q_proj": { "bits": 16 }, "model.layers.3.self_attn.k_proj": { "bits": 16 }, "model.layers.3.self_attn.v_proj": { "bits": 16 }, "model.layers.3.self_attn.o_proj": { "bits": 16 }, "model.layers.3.mlp.router.router": { "bits": 16 }, "model.layers.4.self_attn.q_proj": { "bits": 16 }, "model.layers.4.self_attn.k_proj": { "bits": 16 }, "model.layers.4.self_attn.v_proj": { "bits": 16 }, "model.layers.4.self_attn.o_proj": { "bits": 16 }, "model.layers.4.mlp.router.router": { "bits": 16 }, "model.layers.5.self_attn.q_proj": { "bits": 16 }, "model.layers.5.self_attn.k_proj": { "bits": 16 }, "model.layers.5.self_attn.v_proj": { "bits": 16 }, "model.layers.5.self_attn.o_proj": { "bits": 16 }, "model.layers.5.mlp.router.router": { "bits": 16 }, "model.layers.6.self_attn.q_proj": { "bits": 16 }, "model.layers.6.self_attn.k_proj": { "bits": 16 }, "model.layers.6.self_attn.v_proj": { "bits": 16 }, "model.layers.6.self_attn.o_proj": { "bits": 16 }, "model.layers.6.mlp.router.router": { "bits": 16 }, "model.layers.7.self_attn.q_proj": { "bits": 16 }, "model.layers.7.self_attn.k_proj": { "bits": 16 }, "model.layers.7.self_attn.v_proj": { "bits": 16 }, "model.layers.7.self_attn.o_proj": { "bits": 16 }, "model.layers.7.mlp.router.router": { "bits": 16 }, "model.layers.8.self_attn.q_proj": { "bits": 16 }, "model.layers.8.self_attn.k_proj": { "bits": 16 }, "model.layers.8.self_attn.v_proj": { "bits": 16 }, "model.layers.8.self_attn.o_proj": { "bits": 16 }, "model.layers.8.mlp.router.router": { "bits": 16 }, "model.layers.9.self_attn.q_proj": { "bits": 16 }, "model.layers.9.self_attn.k_proj": { "bits": 16 }, "model.layers.9.self_attn.v_proj": { "bits": 16 }, "model.layers.9.self_attn.o_proj": { "bits": 16 }, "model.layers.9.mlp.router.router": { "bits": 16 }, "model.layers.10.self_attn.q_proj": { "bits": 16 }, "model.layers.10.self_attn.k_proj": { "bits": 16 }, "model.layers.10.self_attn.v_proj": { "bits": 16 }, "model.layers.10.self_attn.o_proj": { "bits": 16 }, "model.layers.10.mlp.router.router": { "bits": 16 }, "model.layers.11.self_attn.q_proj": { "bits": 16 }, "model.layers.11.self_attn.k_proj": { "bits": 16 }, "model.layers.11.self_attn.v_proj": { "bits": 16 }, "model.layers.11.self_attn.o_proj": { "bits": 16 }, "model.layers.11.mlp.router.router": { "bits": 16 }, "model.layers.12.self_attn.q_proj": { "bits": 16 }, "model.layers.12.self_attn.k_proj": { "bits": 16 }, "model.layers.12.self_attn.v_proj": { "bits": 16 }, "model.layers.12.self_attn.o_proj": { "bits": 16 }, "model.layers.12.mlp.router.router": { "bits": 16 }, "model.layers.13.self_attn.q_proj": { "bits": 16 }, "model.layers.13.self_attn.k_proj": { "bits": 16 }, "model.layers.13.self_attn.v_proj": { "bits": 16 }, "model.layers.13.self_attn.o_proj": { "bits": 16 }, "model.layers.13.mlp.router.router": { "bits": 16 }, "model.layers.14.self_attn.q_proj": { "bits": 16 }, "model.layers.14.self_attn.k_proj": { "bits": 16 }, "model.layers.14.self_attn.v_proj": { "bits": 16 }, "model.layers.14.self_attn.o_proj": { "bits": 16 }, "model.layers.14.mlp.router.router": { "bits": 16 }, "model.layers.15.self_attn.q_proj": { "bits": 16 }, "model.layers.15.self_attn.k_proj": { "bits": 16 }, "model.layers.15.self_attn.v_proj": { "bits": 16 }, "model.layers.15.self_attn.o_proj": { "bits": 16 }, "model.layers.15.mlp.router.router": { "bits": 16 }, "model.layers.16.self_attn.q_proj": { "bits": 16 }, "model.layers.16.self_attn.k_proj": { "bits": 16 }, "model.layers.16.self_attn.v_proj": { "bits": 16 }, "model.layers.16.self_attn.o_proj": { "bits": 16 }, "model.layers.16.mlp.router.router": { "bits": 16 }, "model.layers.17.self_attn.q_proj": { "bits": 16 }, "model.layers.17.self_attn.k_proj": { "bits": 16 }, "model.layers.17.self_attn.v_proj": { "bits": 16 }, "model.layers.17.self_attn.o_proj": { "bits": 16 }, "model.layers.17.mlp.router.router": { "bits": 16 }, "model.layers.18.self_attn.q_proj": { "bits": 16 }, "model.layers.18.self_attn.k_proj": { "bits": 16 }, "model.layers.18.self_attn.v_proj": { "bits": 16 }, "model.layers.18.self_attn.o_proj": { "bits": 16 }, "model.layers.18.mlp.router.router": { "bits": 16 }, "model.layers.19.self_attn.q_proj": { "bits": 16 }, "model.layers.19.self_attn.k_proj": { "bits": 16 }, "model.layers.19.self_attn.v_proj": { "bits": 16 }, "model.layers.19.self_attn.o_proj": { "bits": 16 }, "model.layers.19.mlp.router.router": { "bits": 16 }, "model.layers.20.self_attn.q_proj": { "bits": 16 }, "model.layers.20.self_attn.k_proj": { "bits": 16 }, "model.layers.20.self_attn.v_proj": { "bits": 16 }, "model.layers.20.self_attn.o_proj": { "bits": 16 }, "model.layers.20.mlp.router.router": { "bits": 16 }, "model.layers.21.self_attn.q_proj": { "bits": 16 }, "model.layers.21.self_attn.k_proj": { "bits": 16 }, "model.layers.21.self_attn.v_proj": { "bits": 16 }, "model.layers.21.self_attn.o_proj": { "bits": 16 }, "model.layers.21.mlp.router.router": { "bits": 16 }, "model.layers.22.self_attn.q_proj": { "bits": 16 }, "model.layers.22.self_attn.k_proj": { "bits": 16 }, "model.layers.22.self_attn.v_proj": { "bits": 16 }, "model.layers.22.self_attn.o_proj": { "bits": 16 }, "model.layers.22.mlp.router.router": { "bits": 16 }, "model.layers.23.self_attn.q_proj": { "bits": 16 }, "model.layers.23.self_attn.k_proj": { "bits": 16 }, "model.layers.23.self_attn.v_proj": { "bits": 16 }, "model.layers.23.self_attn.o_proj": { "bits": 16 }, "model.layers.23.mlp.router.router": { "bits": 16 }, "model.layers.24.self_attn.q_proj": { "bits": 16 }, "model.layers.24.self_attn.k_proj": { "bits": 16 }, "model.layers.24.self_attn.v_proj": { "bits": 16 }, "model.layers.24.self_attn.o_proj": { "bits": 16 }, "model.layers.24.mlp.router.router": { "bits": 16 }, "model.layers.25.self_attn.q_proj": { "bits": 16 }, "model.layers.25.self_attn.k_proj": { "bits": 16 }, "model.layers.25.self_attn.v_proj": { "bits": 16 }, "model.layers.25.self_attn.o_proj": { "bits": 16 }, "model.layers.25.mlp.router.router": { "bits": 16 }, "model.layers.26.self_attn.q_proj": { "bits": 16 }, "model.layers.26.self_attn.k_proj": { "bits": 16 }, "model.layers.26.self_attn.v_proj": { "bits": 16 }, "model.layers.26.self_attn.o_proj": { "bits": 16 }, "model.layers.26.mlp.router.router": { "bits": 16 }, "model.layers.27.self_attn.q_proj": { "bits": 16 }, "model.layers.27.self_attn.k_proj": { "bits": 16 }, "model.layers.27.self_attn.v_proj": { "bits": 16 }, "model.layers.27.self_attn.o_proj": { "bits": 16 }, "model.layers.27.mlp.router.router": { "bits": 16 }, "model.layers.28.self_attn.q_proj": { "bits": 16 }, "model.layers.28.self_attn.k_proj": { "bits": 16 }, "model.layers.28.self_attn.v_proj": { "bits": 16 }, "model.layers.28.self_attn.o_proj": { "bits": 16 }, "model.layers.28.mlp.router.router": { "bits": 16 }, "model.layers.29.self_attn.q_proj": { "bits": 16 }, "model.layers.29.self_attn.k_proj": { "bits": 16 }, "model.layers.29.self_attn.v_proj": { "bits": 16 }, "model.layers.29.self_attn.o_proj": { "bits": 16 }, "model.layers.29.mlp.router.router": { "bits": 16 }, "model.layers.30.self_attn.q_proj": { "bits": 16 }, "model.layers.30.self_attn.k_proj": { "bits": 16 }, "model.layers.30.self_attn.v_proj": { "bits": 16 }, "model.layers.30.self_attn.o_proj": { "bits": 16 }, "model.layers.30.mlp.router.router": { "bits": 16 }, "model.layers.31.self_attn.q_proj": { "bits": 16 }, "model.layers.31.self_attn.k_proj": { "bits": 16 }, "model.layers.31.self_attn.v_proj": { "bits": 16 }, "model.layers.31.self_attn.o_proj": { "bits": 16 }, "model.layers.31.mlp.router.router": { "bits": 16 }, "model.layers.32.self_attn.q_proj": { "bits": 16 }, "model.layers.32.self_attn.k_proj": { "bits": 16 }, "model.layers.32.self_attn.v_proj": { "bits": 16 }, "model.layers.32.self_attn.o_proj": { "bits": 16 }, "model.layers.32.mlp.router.router": { "bits": 16 }, "model.layers.33.self_attn.q_proj": { "bits": 16 }, "model.layers.33.self_attn.k_proj": { "bits": 16 }, "model.layers.33.self_attn.v_proj": { "bits": 16 }, "model.layers.33.self_attn.o_proj": { "bits": 16 }, "model.layers.33.mlp.router.router": { "bits": 16 }, "model.layers.34.self_attn.q_proj": { "bits": 16 }, "model.layers.34.self_attn.k_proj": { "bits": 16 }, "model.layers.34.self_attn.v_proj": { "bits": 16 }, "model.layers.34.self_attn.o_proj": { "bits": 16 }, "model.layers.34.mlp.router.router": { "bits": 16 }, "model.layers.35.self_attn.q_proj": { "bits": 16 }, "model.layers.35.self_attn.k_proj": { "bits": 16 }, "model.layers.35.self_attn.v_proj": { "bits": 16 }, "model.layers.35.self_attn.o_proj": { "bits": 16 }, "model.layers.35.mlp.router.router": { "bits": 16 } } }