{ "architectures": [ "DeepseekV3ForCausalLM" ], "attention_bias": false, "attention_dropout": 0.0, "auto_map_bak": { "AutoConfig": "configuration_deepseek.DeepseekV3Config", "AutoModel": "modeling_deepseek.DeepseekV3Model", "AutoModelForCausalLM": "modeling_deepseek.DeepseekV3ForCausalLM" }, "bos_token_id": 0, "eos_token_id": 1, "ep_size": 1, "first_k_dense_replace": 3, "head_dim": 64, "hidden_act": "silu", "hidden_size": 7168, "initializer_range": 0.02, "intermediate_size": 18432, "kv_lora_rank": 512, "max_position_embeddings": 163840, "model_type": "deepseek_v3", "moe_intermediate_size": 2048, "moe_layer_freq": 1, "n_group": 8, "n_routed_experts": 256, "n_shared_experts": 1, "norm_topk_prob": true, "num_attention_heads": 128, "num_experts_per_tok": 8, "num_hidden_layers": 61, "num_key_value_heads": 128, "num_nextn_predict_layers": 1, "pretraining_tp": 1, "q_lora_rank": 1536, "qk_head_dim": 192, "qk_nope_head_dim": 128, "qk_rope_head_dim": 64, "quantization_config": { "autoround_version": "0.5.1", "batch_size": 4, "bits": 2, "data_type": "int", "extra_config": { "model.layers.0.mlp.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.0.mlp.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.0.mlp.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.0.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.0.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.0.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.0.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.0.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.1.mlp.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.1.mlp.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.1.mlp.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.1.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.1.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.1.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.1.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.1.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.10.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.10.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.10.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.10.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.10.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.10.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.10.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.10.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.11.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.11.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.11.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.11.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.11.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.11.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.11.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.11.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.12.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.12.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.12.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.12.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.12.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.12.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.12.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.12.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.13.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.13.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.13.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.13.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.13.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.13.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.13.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.13.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.14.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.14.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.14.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.14.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.14.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.14.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.14.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.14.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.15.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.15.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.15.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.15.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.15.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.15.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.15.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.15.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.16.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.16.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.16.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.16.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.16.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.16.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.16.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.16.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.17.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.17.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.17.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.17.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.17.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.17.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.17.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.17.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.18.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.18.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.18.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.18.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.18.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.18.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.18.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.18.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.19.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.19.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.19.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.19.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.19.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.19.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.19.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.19.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.2.mlp.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.2.mlp.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.2.mlp.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.2.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.2.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.2.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.2.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.2.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.20.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.20.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.20.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.20.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.20.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.20.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.20.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.20.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.21.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.21.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.21.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.21.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.21.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.21.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.21.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.21.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.22.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.22.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.22.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.22.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.22.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.22.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.22.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.22.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.23.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.23.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.23.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.23.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.23.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.23.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.23.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.23.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.24.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.24.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.24.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.24.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.24.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.24.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.24.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.24.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.25.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.25.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.25.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.25.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.25.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.25.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.25.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.25.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.26.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.26.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.26.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.26.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.26.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.26.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.26.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.26.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.27.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.27.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.27.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.27.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.27.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.27.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.27.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.27.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.28.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.28.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.28.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.28.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.28.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.28.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.28.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.28.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.29.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.29.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.29.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.29.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.29.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.29.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.29.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.29.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.3.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.3.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.3.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.3.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.3.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.3.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.3.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.3.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.30.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.30.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.30.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.30.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.30.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.30.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.30.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.30.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.31.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.31.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.31.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.31.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.31.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.31.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.31.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.31.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.32.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.32.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.32.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.32.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.32.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.32.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.32.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.32.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.33.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.33.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.33.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.33.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.33.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.33.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.33.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.33.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.34.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.34.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.34.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.34.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.34.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.34.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.34.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.34.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.35.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.35.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.35.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.35.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.35.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.35.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.35.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.35.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.36.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.36.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.36.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.36.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.36.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.36.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.36.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.36.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.37.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.37.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.37.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.37.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.37.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.37.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.37.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.37.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.38.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.38.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.38.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.38.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.38.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.38.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.38.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.38.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.39.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.39.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.39.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.39.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.39.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.39.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.39.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.39.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.4.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.4.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.4.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.4.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.4.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.4.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.4.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.4.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.40.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.40.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.40.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.40.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.40.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.40.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.40.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.40.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.41.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.41.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.41.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.41.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.41.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.41.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.41.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.41.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.42.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.42.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.42.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.42.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.42.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.42.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.42.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.42.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.43.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.43.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.43.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.43.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.43.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.43.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.43.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.43.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.44.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.44.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.44.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.44.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.44.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.44.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.44.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.44.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.45.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.45.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.45.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.45.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.45.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.45.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.45.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.45.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.46.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.46.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.46.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.46.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.46.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.46.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.46.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.46.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.47.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.47.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.47.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.47.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.47.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.47.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.47.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.47.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.48.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.48.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.48.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.48.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.48.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.48.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.48.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.48.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.49.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.49.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.49.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.49.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.49.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.49.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.49.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.49.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.5.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.5.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.5.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.5.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.5.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.5.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.5.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.5.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.50.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.50.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.50.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.50.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.50.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.50.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.50.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.50.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.51.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.51.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.51.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.51.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.51.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.51.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.51.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.51.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.52.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.52.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.52.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.52.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.52.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.52.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.52.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.52.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.53.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.53.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.53.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.53.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.53.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.53.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.53.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.53.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.54.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.54.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.54.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.54.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.54.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.54.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.54.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.54.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.55.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.55.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.55.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.55.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.55.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.55.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.55.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.55.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.56.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.56.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.56.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.56.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.56.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.56.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.56.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.56.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.57.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.57.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.57.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.57.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.57.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.57.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.57.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.57.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.58.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.58.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.58.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.58.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.58.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.58.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.58.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.58.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.59.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.59.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.59.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.59.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.59.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.59.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.59.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.59.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.6.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.6.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.6.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.6.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.6.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.6.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.6.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.6.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.60.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.60.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.60.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.60.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.60.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.60.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.60.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.60.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.7.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.7.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.7.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.7.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.7.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.7.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.7.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.7.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.8.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.8.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.8.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.8.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.8.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.8.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.8.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.8.self_attn.q_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.9.mlp.shared_experts.down_proj": { "bits": 4, "group_size": 128 }, "model.layers.9.mlp.shared_experts.gate_proj": { "bits": 4, "group_size": 128 }, "model.layers.9.mlp.shared_experts.up_proj": { "bits": 4, "group_size": 128 }, "model.layers.9.self_attn.kv_a_proj_with_mqa": { "bits": 4, "group_size": 128 }, "model.layers.9.self_attn.kv_b_proj": { "bits": 4, "group_size": 128 }, "model.layers.9.self_attn.o_proj": { "bits": 4, "group_size": 128 }, "model.layers.9.self_attn.q_a_proj": { "bits": 4, "group_size": 128 }, "model.layers.9.self_attn.q_b_proj": { "bits": 4, "group_size": 128 } }, "group_size": 64, "iters": 400, "nsamples": 512, "packing_format": "auto_round:auto_gptq", "quant_method": "auto-round", "seqlen": 512, "sym": true }, "quantization_config_bak": { "activation_scheme": "dynamic", "fmt": "e4m3", "quant_method": "fp8", "weight_block_size": [ 128, 128 ] }, "rms_norm_eps": 1e-06, "rope_interleave": true, "rope_scaling": { "beta_fast": 32.0, "beta_slow": 1.0, "factor": 40.0, "mscale": 1.0, "mscale_all_dim": 1.0, "original_max_position_embeddings": 4096, "rope_type": "yarn", "type": "yarn" }, "rope_theta": 10000, "routed_scaling_factor": 2.5, "scoring_func": "sigmoid", "tie_word_embeddings": false, "topk_group": 4, "topk_method": "noaux_tc", "torch_dtype": "bfloat16", "transformers_version": "4.52.3", "use_cache": true, "v_head_dim": 128, "vocab_size": 129280 }