{ "architectures": [ "DeepseekV3ForCausalLM" ], "attention_bias": false, "attention_dropout": 0.0, "auto_map": { "AutoConfig": "configuration_deepseek.DeepseekV3Config", "AutoModel": "modeling_deepseek.DeepseekV3Model", "AutoModelForCausalLM": "modeling_deepseek.DeepseekV3ForCausalLM" }, "aux_loss_alpha": 0.001, "bos_token_id": 0, "eos_token_id": 1, "ep_size": 1, "first_k_dense_replace": 3, "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, "pad_token_id": 128815, "pretraining_tp": 1, "q_lora_rank": 1536, "qk_nope_head_dim": 128, "qk_rope_head_dim": 64, "quantization_config": { "algo_config": { "compute_scale_loss": "MAE", "model_decoder_layers": "model.layers", "name": "autosmoothquant", "num_attention_heads": -1, "num_key_value_heads": -1, "scaling_layers": [ { "inp": "mlp.gate_proj", "layers": [ "mlp.gate_proj", "mlp.up_proj" ], "module2inspect": "mlp", "prev_op": "post_attention_layernorm" }, { "inp": "mlp.down_proj", "layers": [ "mlp.down_proj" ], "prev_op": "mlp.up_proj" }, { "inp": "down_proj", "layers": [ "down_proj" ], "prev_op": "up_proj" } ] }, "exclude": [ "model.layers.0.self_attn.q_a_proj", "model.layers.0.self_attn.q_b_proj", "model.layers.0.self_attn.kv_a_proj_with_mqa", "model.layers.0.self_attn.kv_b_proj", "model.layers.0.self_attn.o_proj", "model.layers.1.self_attn.q_a_proj", "model.layers.1.self_attn.q_b_proj", "model.layers.1.self_attn.kv_a_proj_with_mqa", "model.layers.1.self_attn.kv_b_proj", "model.layers.1.self_attn.o_proj", "model.layers.2.self_attn.q_a_proj", "model.layers.2.self_attn.q_b_proj", "model.layers.2.self_attn.kv_a_proj_with_mqa", "model.layers.2.self_attn.kv_b_proj", "model.layers.2.self_attn.o_proj", "model.layers.3.self_attn.q_a_proj", "model.layers.3.self_attn.q_b_proj", "model.layers.3.self_attn.kv_a_proj_with_mqa", "model.layers.3.self_attn.kv_b_proj", "model.layers.3.self_attn.o_proj", "model.layers.4.self_attn.q_a_proj", "model.layers.4.self_attn.q_b_proj", "model.layers.4.self_attn.kv_a_proj_with_mqa", "model.layers.4.self_attn.kv_b_proj", "model.layers.4.self_attn.o_proj", "model.layers.5.self_attn.q_a_proj", "model.layers.5.self_attn.q_b_proj", "model.layers.5.self_attn.kv_a_proj_with_mqa", "model.layers.5.self_attn.kv_b_proj", "model.layers.5.self_attn.o_proj", "model.layers.6.self_attn.q_a_proj", "model.layers.6.self_attn.q_b_proj", "model.layers.6.self_attn.kv_a_proj_with_mqa", "model.layers.6.self_attn.kv_b_proj", "model.layers.6.self_attn.o_proj", "model.layers.7.self_attn.q_a_proj", "model.layers.7.self_attn.q_b_proj", "model.layers.7.self_attn.kv_a_proj_with_mqa", "model.layers.7.self_attn.kv_b_proj", "model.layers.7.self_attn.o_proj", "model.layers.8.self_attn.q_a_proj", "model.layers.8.self_attn.q_b_proj", "model.layers.8.self_attn.kv_a_proj_with_mqa", "model.layers.8.self_attn.kv_b_proj", "model.layers.8.self_attn.o_proj", "model.layers.9.self_attn.q_a_proj", "model.layers.9.self_attn.q_b_proj", "model.layers.9.self_attn.kv_a_proj_with_mqa", "model.layers.9.self_attn.kv_b_proj", "model.layers.9.self_attn.o_proj", "model.layers.10.self_attn.q_a_proj", "model.layers.10.self_attn.q_b_proj", "model.layers.10.self_attn.kv_a_proj_with_mqa", "model.layers.10.self_attn.kv_b_proj", "model.layers.10.self_attn.o_proj", "model.layers.11.self_attn.q_a_proj", "model.layers.11.self_attn.q_b_proj", "model.layers.11.self_attn.kv_a_proj_with_mqa", "model.layers.11.self_attn.kv_b_proj", "model.layers.11.self_attn.o_proj", "model.layers.12.self_attn.q_a_proj", "model.layers.12.self_attn.q_b_proj", "model.layers.12.self_attn.kv_a_proj_with_mqa", "model.layers.12.self_attn.kv_b_proj", "model.layers.12.self_attn.o_proj", "model.layers.13.self_attn.q_a_proj", "model.layers.13.self_attn.q_b_proj", "model.layers.13.self_attn.kv_a_proj_with_mqa", "model.layers.13.self_attn.kv_b_proj", "model.layers.13.self_attn.o_proj", "model.layers.14.self_attn.q_a_proj", "model.layers.14.self_attn.q_b_proj", "model.layers.14.self_attn.kv_a_proj_with_mqa", "model.layers.14.self_attn.kv_b_proj", "model.layers.14.self_attn.o_proj", "model.layers.15.self_attn.q_a_proj", "model.layers.15.self_attn.q_b_proj", "model.layers.15.self_attn.kv_a_proj_with_mqa", "model.layers.15.self_attn.kv_b_proj", "model.layers.15.self_attn.o_proj", "model.layers.16.self_attn.q_a_proj", "model.layers.16.self_attn.q_b_proj", "model.layers.16.self_attn.kv_a_proj_with_mqa", "model.layers.16.self_attn.kv_b_proj", "model.layers.16.self_attn.o_proj", "model.layers.17.self_attn.q_a_proj", "model.layers.17.self_attn.q_b_proj", "model.layers.17.self_attn.kv_a_proj_with_mqa", "model.layers.17.self_attn.kv_b_proj", "model.layers.17.self_attn.o_proj", "model.layers.18.self_attn.q_a_proj", "model.layers.18.self_attn.q_b_proj", "model.layers.18.self_attn.kv_a_proj_with_mqa", "model.layers.18.self_attn.kv_b_proj", "model.layers.18.self_attn.o_proj", "model.layers.19.self_attn.q_a_proj", "model.layers.19.self_attn.q_b_proj", "model.layers.19.self_attn.kv_a_proj_with_mqa", "model.layers.19.self_attn.kv_b_proj", "model.layers.19.self_attn.o_proj", "model.layers.20.self_attn.q_a_proj", "model.layers.20.self_attn.q_b_proj", "model.layers.20.self_attn.kv_a_proj_with_mqa", "model.layers.20.self_attn.kv_b_proj", "model.layers.20.self_attn.o_proj", "model.layers.21.self_attn.q_a_proj", "model.layers.21.self_attn.q_b_proj", "model.layers.21.self_attn.kv_a_proj_with_mqa", "model.layers.21.self_attn.kv_b_proj", "model.layers.21.self_attn.o_proj", "model.layers.22.self_attn.q_a_proj", "model.layers.22.self_attn.q_b_proj", "model.layers.22.self_attn.kv_a_proj_with_mqa", "model.layers.22.self_attn.kv_b_proj", "model.layers.22.self_attn.o_proj", "model.layers.23.self_attn.q_a_proj", "model.layers.23.self_attn.q_b_proj", "model.layers.23.self_attn.kv_a_proj_with_mqa", "model.layers.23.self_attn.kv_b_proj", "model.layers.23.self_attn.o_proj", "model.layers.24.self_attn.q_a_proj", "model.layers.24.self_attn.q_b_proj", "model.layers.24.self_attn.kv_a_proj_with_mqa", "model.layers.24.self_attn.kv_b_proj", "model.layers.24.self_attn.o_proj", "model.layers.25.self_attn.q_a_proj", "model.layers.25.self_attn.q_b_proj", "model.layers.25.self_attn.kv_a_proj_with_mqa", "model.layers.25.self_attn.kv_b_proj", "model.layers.25.self_attn.o_proj", "model.layers.26.self_attn.q_a_proj", "model.layers.26.self_attn.q_b_proj", "model.layers.26.self_attn.kv_a_proj_with_mqa", "model.layers.26.self_attn.kv_b_proj", "model.layers.26.self_attn.o_proj", "model.layers.27.self_attn.q_a_proj", "model.layers.27.self_attn.q_b_proj", "model.layers.27.self_attn.kv_a_proj_with_mqa", "model.layers.27.self_attn.kv_b_proj", "model.layers.27.self_attn.o_proj", "model.layers.28.self_attn.q_a_proj", "model.layers.28.self_attn.q_b_proj", "model.layers.28.self_attn.kv_a_proj_with_mqa", "model.layers.28.self_attn.kv_b_proj", "model.layers.28.self_attn.o_proj", "model.layers.29.self_attn.q_a_proj", "model.layers.29.self_attn.q_b_proj", "model.layers.29.self_attn.kv_a_proj_with_mqa", "model.layers.29.self_attn.kv_b_proj", "model.layers.29.self_attn.o_proj", "model.layers.30.self_attn.q_a_proj", "model.layers.30.self_attn.q_b_proj", "model.layers.30.self_attn.kv_a_proj_with_mqa", "model.layers.30.self_attn.kv_b_proj", "model.layers.30.self_attn.o_proj", "model.layers.31.self_attn.q_a_proj", "model.layers.31.self_attn.q_b_proj", "model.layers.31.self_attn.kv_a_proj_with_mqa", "model.layers.31.self_attn.kv_b_proj", "model.layers.31.self_attn.o_proj", "model.layers.32.self_attn.q_a_proj", "model.layers.32.self_attn.q_b_proj", "model.layers.32.self_attn.kv_a_proj_with_mqa", "model.layers.32.self_attn.kv_b_proj", "model.layers.32.self_attn.o_proj", "model.layers.33.self_attn.q_a_proj", "model.layers.33.self_attn.q_b_proj", "model.layers.33.self_attn.kv_a_proj_with_mqa", "model.layers.33.self_attn.kv_b_proj", "model.layers.33.self_attn.o_proj", "model.layers.34.self_attn.q_a_proj", "model.layers.34.self_attn.q_b_proj", "model.layers.34.self_attn.kv_a_proj_with_mqa", "model.layers.34.self_attn.kv_b_proj", "model.layers.34.self_attn.o_proj", "model.layers.35.self_attn.q_a_proj", "model.layers.35.self_attn.q_b_proj", "model.layers.35.self_attn.kv_a_proj_with_mqa", "model.layers.35.self_attn.kv_b_proj", "model.layers.35.self_attn.o_proj", "model.layers.36.self_attn.q_a_proj", "model.layers.36.self_attn.q_b_proj", "model.layers.36.self_attn.kv_a_proj_with_mqa", "model.layers.36.self_attn.kv_b_proj", "model.layers.36.self_attn.o_proj", "model.layers.37.self_attn.q_a_proj", "model.layers.37.self_attn.q_b_proj", "model.layers.37.self_attn.kv_a_proj_with_mqa", "model.layers.37.self_attn.kv_b_proj", "model.layers.37.self_attn.o_proj", "model.layers.38.self_attn.q_a_proj", "model.layers.38.self_attn.q_b_proj", "model.layers.38.self_attn.kv_a_proj_with_mqa", "model.layers.38.self_attn.kv_b_proj", "model.layers.38.self_attn.o_proj", "model.layers.39.self_attn.q_a_proj", "model.layers.39.self_attn.q_b_proj", "model.layers.39.self_attn.kv_a_proj_with_mqa", "model.layers.39.self_attn.kv_b_proj", "model.layers.39.self_attn.o_proj", "model.layers.40.self_attn.q_a_proj", "model.layers.40.self_attn.q_b_proj", "model.layers.40.self_attn.kv_a_proj_with_mqa", "model.layers.40.self_attn.kv_b_proj", "model.layers.40.self_attn.o_proj", "model.layers.41.self_attn.q_a_proj", "model.layers.41.self_attn.q_b_proj", "model.layers.41.self_attn.kv_a_proj_with_mqa", "model.layers.41.self_attn.kv_b_proj", "model.layers.41.self_attn.o_proj", "model.layers.42.self_attn.q_a_proj", "model.layers.42.self_attn.q_b_proj", "model.layers.42.self_attn.kv_a_proj_with_mqa", "model.layers.42.self_attn.kv_b_proj", "model.layers.42.self_attn.o_proj", "model.layers.43.self_attn.q_a_proj", "model.layers.43.self_attn.q_b_proj", "model.layers.43.self_attn.kv_a_proj_with_mqa", "model.layers.43.self_attn.kv_b_proj", "model.layers.43.self_attn.o_proj", "model.layers.44.self_attn.q_a_proj", "model.layers.44.self_attn.q_b_proj", "model.layers.44.self_attn.kv_a_proj_with_mqa", "model.layers.44.self_attn.kv_b_proj", "model.layers.44.self_attn.o_proj", "model.layers.45.self_attn.q_a_proj", "model.layers.45.self_attn.q_b_proj", "model.layers.45.self_attn.kv_a_proj_with_mqa", "model.layers.45.self_attn.kv_b_proj", "model.layers.45.self_attn.o_proj", "model.layers.46.self_attn.q_a_proj", "model.layers.46.self_attn.q_b_proj", "model.layers.46.self_attn.kv_a_proj_with_mqa", "model.layers.46.self_attn.kv_b_proj", "model.layers.46.self_attn.o_proj", "model.layers.47.self_attn.q_a_proj", "model.layers.47.self_attn.q_b_proj", "model.layers.47.self_attn.kv_a_proj_with_mqa", "model.layers.47.self_attn.kv_b_proj", "model.layers.47.self_attn.o_proj", "model.layers.48.self_attn.q_a_proj", "model.layers.48.self_attn.q_b_proj", "model.layers.48.self_attn.kv_a_proj_with_mqa", "model.layers.48.self_attn.kv_b_proj", "model.layers.48.self_attn.o_proj", "model.layers.49.self_attn.q_a_proj", "model.layers.49.self_attn.q_b_proj", "model.layers.49.self_attn.kv_a_proj_with_mqa", "model.layers.49.self_attn.kv_b_proj", "model.layers.49.self_attn.o_proj", "model.layers.50.self_attn.q_a_proj", "model.layers.50.self_attn.q_b_proj", "model.layers.50.self_attn.kv_a_proj_with_mqa", "model.layers.50.self_attn.kv_b_proj", "model.layers.50.self_attn.o_proj", "model.layers.51.self_attn.q_a_proj", "model.layers.51.self_attn.q_b_proj", "model.layers.51.self_attn.kv_a_proj_with_mqa", "model.layers.51.self_attn.kv_b_proj", "model.layers.51.self_attn.o_proj", "model.layers.52.self_attn.q_a_proj", "model.layers.52.self_attn.q_b_proj", "model.layers.52.self_attn.kv_a_proj_with_mqa", "model.layers.52.self_attn.kv_b_proj", "model.layers.52.self_attn.o_proj", "model.layers.53.self_attn.q_a_proj", "model.layers.53.self_attn.q_b_proj", "model.layers.53.self_attn.kv_a_proj_with_mqa", "model.layers.53.self_attn.kv_b_proj", "model.layers.53.self_attn.o_proj", "model.layers.54.self_attn.q_a_proj", "model.layers.54.self_attn.q_b_proj", "model.layers.54.self_attn.kv_a_proj_with_mqa", "model.layers.54.self_attn.kv_b_proj", "model.layers.54.self_attn.o_proj", "model.layers.55.self_attn.q_a_proj", "model.layers.55.self_attn.q_b_proj", "model.layers.55.self_attn.kv_a_proj_with_mqa", "model.layers.55.self_attn.kv_b_proj", "model.layers.55.self_attn.o_proj", "model.layers.56.self_attn.q_a_proj", "model.layers.56.self_attn.q_b_proj", "model.layers.56.self_attn.kv_a_proj_with_mqa", "model.layers.56.self_attn.kv_b_proj", "model.layers.56.self_attn.o_proj", "model.layers.57.self_attn.q_a_proj", "model.layers.57.self_attn.q_b_proj", "model.layers.57.self_attn.kv_a_proj_with_mqa", "model.layers.57.self_attn.kv_b_proj", "model.layers.57.self_attn.o_proj", "model.layers.58.self_attn.q_a_proj", "model.layers.58.self_attn.q_b_proj", "model.layers.58.self_attn.kv_a_proj_with_mqa", "model.layers.58.self_attn.kv_b_proj", "model.layers.58.self_attn.o_proj", "model.layers.59.self_attn.q_a_proj", "model.layers.59.self_attn.q_b_proj", "model.layers.59.self_attn.kv_a_proj_with_mqa", "model.layers.59.self_attn.kv_b_proj", "model.layers.59.self_attn.o_proj", "model.layers.60.self_attn.q_a_proj", "model.layers.60.self_attn.q_b_proj", "model.layers.60.self_attn.kv_a_proj_with_mqa", "model.layers.60.self_attn.kv_b_proj", "model.layers.60.self_attn.o_proj", "lm_head" ], "export": { "kv_cache_group": [], "min_kv_scale": 0.0, "pack_method": "reorder", "weight_format": "real_quantized", "weight_merge_groups": null }, "global_quant_config": { "bias": null, "input_tensors": { "ch_axis": -1, "dtype": "fp4", "group_size": 32, "is_dynamic": true, "is_mx_scale_constraint": false, "is_scale_quant": false, "mx_element_dtype": null, "observer_cls": "PerBlockMXObserver", "qscheme": "per_group", "round_method": "half_even", "scale_calculation_mode": "even", "scale_format": "e8m0", "scale_type": "float", "symmetric": null }, "output_tensors": null, "target_device": null, "weight": { "ch_axis": -1, "dtype": "fp4", "group_size": 32, "is_dynamic": false, "is_mx_scale_constraint": false, "is_scale_quant": false, "mx_element_dtype": null, "observer_cls": "PerBlockMXObserver", "qscheme": "per_group", "round_method": "half_even", "scale_calculation_mode": "even", "scale_format": "e8m0", "scale_type": "float", "symmetric": null } }, "layer_quant_config": {}, "layer_type_quant_config": {}, "quant_method": "quark", "quant_mode": "eager_mode", "softmax_quant_spec": null }, "rms_norm_eps": 1e-06, "rope_scaling": { "beta_fast": 32, "beta_slow": 1, "factor": 40, "mscale": 1.0, "mscale_all_dim": 1.0, "original_max_position_embeddings": 4096, "type": "yarn" }, "rope_theta": 10000, "routed_scaling_factor": 2.5, "scoring_func": "sigmoid", "seq_aux": true, "tie_word_embeddings": false, "topk_group": 4, "topk_method": "noaux_tc", "torch_dtype": "bfloat16", "transformers_version": "4.51.3", "unsloth_fixed": true, "use_cache": true, "v_head_dim": 128, "vocab_size": 129280 }