{ "architectures": [ "DeepseekV4ForCausalLM" ], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 0, "eos_token_id": [ 1, 128803 ], "expert_dtype": "fp4", "hc_eps": 1e-06, "hc_mult": 4, "hc_sinkhorn_iters": 20, "head_dim": 512, "hidden_act": "silu", "hidden_size": 4096, "index_head_dim": 128, "index_n_heads": 64, "index_topk": 512, "initializer_range": 0.02, "max_position_embeddings": 1048576, "model_type": "deepseek_v4", "moe_intermediate_size": 2048, "n_routed_experts": 256, "n_shared_experts": 1, "norm_topk_prob": true, "num_attention_heads": 64, "num_experts_per_tok": 6, "num_hidden_layers": 43, "num_hash_layers": 3, "num_key_value_heads": 1, "num_nextn_predict_layers": 0, "o_groups": 8, "o_lora_rank": 1024, "q_lora_rank": 1024, "qk_rope_head_dim": 64, "rms_norm_eps": 1e-06, "rope_theta": 10000, "routed_scaling_factor": 1.5, "scoring_func": "sqrtsoftplus", "sliding_window": 128, "swiglu_limit": 10.0, "tie_word_embeddings": false, "topk_method": "noaux_tc", "torch_dtype": "bfloat16", "transformers_version": "4.57.1", "use_cache": true, "vocab_size": 129280, "compress_rope_theta": 160000, "compress_ratios": [ 0, 0, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 128, 4, 0 ], "rope_parameters": { "beta_fast": 32.0, "beta_slow": 1.0, "factor": 16.0, "original_max_position_embeddings": 65536, "rope_type": "yarn", "rope_theta": 10000.0 }, "quantization": { "bits": 8, "group_size": 32, "mode": "affine", "routed_expert_bits": 2, "routed_expert_bit_plan": { "codec": "mxtq", "target": "q3.4", "effective_bits": 3.3953, "bits_used": [ 2, 4 ], "group_count": 129, "bit_counts": { "2": 39, "4": 90 }, "assignment": { "layer_0.down_proj": 4, "layer_0.gate_proj": 4, "layer_0.up_proj": 4, "layer_1.down_proj": 4, "layer_1.gate_proj": 4, "layer_1.up_proj": 4, "layer_2.down_proj": 4, "layer_2.gate_proj": 4, "layer_2.up_proj": 4, "layer_3.down_proj": 4, "layer_3.gate_proj": 2, "layer_3.up_proj": 2, "layer_4.down_proj": 4, "layer_4.gate_proj": 2, "layer_4.up_proj": 2, "layer_5.down_proj": 4, "layer_5.gate_proj": 2, "layer_5.up_proj": 2, "layer_6.down_proj": 4, "layer_6.gate_proj": 2, "layer_6.up_proj": 2, "layer_7.down_proj": 4, "layer_7.gate_proj": 2, "layer_7.up_proj": 2, "layer_8.down_proj": 4, "layer_8.gate_proj": 2, "layer_8.up_proj": 2, "layer_9.down_proj": 4, "layer_9.gate_proj": 2, "layer_9.up_proj": 2, "layer_10.down_proj": 4, "layer_10.gate_proj": 2, "layer_10.up_proj": 2, "layer_11.down_proj": 4, "layer_11.gate_proj": 2, "layer_11.up_proj": 2, "layer_12.down_proj": 4, "layer_12.gate_proj": 2, "layer_12.up_proj": 2, "layer_13.down_proj": 4, "layer_13.gate_proj": 2, "layer_13.up_proj": 2, "layer_14.down_proj": 4, "layer_14.gate_proj": 2, "layer_14.up_proj": 2, "layer_15.down_proj": 4, "layer_15.gate_proj": 2, "layer_15.up_proj": 2, "layer_16.down_proj": 4, "layer_16.gate_proj": 2, "layer_16.up_proj": 2, "layer_17.down_proj": 4, "layer_17.gate_proj": 2, "layer_17.up_proj": 2, "layer_18.down_proj": 4, "layer_18.gate_proj": 2, "layer_18.up_proj": 2, "layer_19.down_proj": 4, "layer_19.gate_proj": 2, "layer_19.up_proj": 2, "layer_20.down_proj": 4, "layer_20.gate_proj": 2, "layer_20.up_proj": 2, "layer_21.down_proj": 4, "layer_21.gate_proj": 2, "layer_21.up_proj": 2, "layer_22.down_proj": 4, "layer_22.gate_proj": 4, "layer_22.up_proj": 2, "layer_23.down_proj": 4, "layer_23.gate_proj": 4, "layer_23.up_proj": 4, "layer_24.down_proj": 4, "layer_24.gate_proj": 4, "layer_24.up_proj": 4, "layer_25.down_proj": 4, "layer_25.gate_proj": 4, "layer_25.up_proj": 4, "layer_26.down_proj": 4, "layer_26.gate_proj": 4, "layer_26.up_proj": 4, "layer_27.down_proj": 4, "layer_27.gate_proj": 4, "layer_27.up_proj": 4, "layer_28.down_proj": 4, "layer_28.gate_proj": 4, "layer_28.up_proj": 4, "layer_29.down_proj": 4, "layer_29.gate_proj": 4, "layer_29.up_proj": 4, "layer_30.down_proj": 4, "layer_30.gate_proj": 4, "layer_30.up_proj": 4, "layer_31.down_proj": 4, "layer_31.gate_proj": 4, "layer_31.up_proj": 4, "layer_32.down_proj": 4, "layer_32.gate_proj": 4, "layer_32.up_proj": 4, "layer_33.down_proj": 4, "layer_33.gate_proj": 4, "layer_33.up_proj": 4, "layer_34.down_proj": 4, "layer_34.gate_proj": 4, "layer_34.up_proj": 4, "layer_35.down_proj": 4, "layer_35.gate_proj": 4, "layer_35.up_proj": 4, "layer_36.down_proj": 4, "layer_36.gate_proj": 4, "layer_36.up_proj": 4, "layer_37.down_proj": 4, "layer_37.gate_proj": 4, "layer_37.up_proj": 4, "layer_38.down_proj": 4, "layer_38.gate_proj": 4, "layer_38.up_proj": 4, "layer_39.down_proj": 4, "layer_39.gate_proj": 4, "layer_39.up_proj": 4, "layer_40.down_proj": 4, "layer_40.gate_proj": 4, "layer_40.up_proj": 4, "layer_41.down_proj": 4, "layer_41.gate_proj": 4, "layer_41.up_proj": 4, "layer_42.down_proj": 4, "layer_42.gate_proj": 4, "layer_42.up_proj": 4 }, "assignment_unit": "layer.projection, shared across all 256 routed experts" }, "mxtq_bits": { "routed_expert": 2, "attention": 8, "shared_expert": 8, "compressor": 8, "indexer": 8, "embed_tokens": 8, "lm_head": 8, "norms_router_hc": 16 } }, "weight_format": "mxtq", "routed_expert_bits": 2, "routed_expert_bit_plan": { "codec": "mxtq", "target": "q3.4", "effective_bits": 3.3953, "bits_used": [ 2, 4 ], "group_count": 129, "bit_counts": { "2": 39, "4": 90 }, "assignment": { "layer_0.down_proj": 4, "layer_0.gate_proj": 4, "layer_0.up_proj": 4, "layer_1.down_proj": 4, "layer_1.gate_proj": 4, "layer_1.up_proj": 4, "layer_2.down_proj": 4, "layer_2.gate_proj": 4, "layer_2.up_proj": 4, "layer_3.down_proj": 4, "layer_3.gate_proj": 2, "layer_3.up_proj": 2, "layer_4.down_proj": 4, "layer_4.gate_proj": 2, "layer_4.up_proj": 2, "layer_5.down_proj": 4, "layer_5.gate_proj": 2, "layer_5.up_proj": 2, "layer_6.down_proj": 4, "layer_6.gate_proj": 2, "layer_6.up_proj": 2, "layer_7.down_proj": 4, "layer_7.gate_proj": 2, "layer_7.up_proj": 2, "layer_8.down_proj": 4, "layer_8.gate_proj": 2, "layer_8.up_proj": 2, "layer_9.down_proj": 4, "layer_9.gate_proj": 2, "layer_9.up_proj": 2, "layer_10.down_proj": 4, "layer_10.gate_proj": 2, "layer_10.up_proj": 2, "layer_11.down_proj": 4, "layer_11.gate_proj": 2, "layer_11.up_proj": 2, "layer_12.down_proj": 4, "layer_12.gate_proj": 2, "layer_12.up_proj": 2, "layer_13.down_proj": 4, "layer_13.gate_proj": 2, "layer_13.up_proj": 2, "layer_14.down_proj": 4, "layer_14.gate_proj": 2, "layer_14.up_proj": 2, "layer_15.down_proj": 4, "layer_15.gate_proj": 2, "layer_15.up_proj": 2, "layer_16.down_proj": 4, "layer_16.gate_proj": 2, "layer_16.up_proj": 2, "layer_17.down_proj": 4, "layer_17.gate_proj": 2, "layer_17.up_proj": 2, "layer_18.down_proj": 4, "layer_18.gate_proj": 2, "layer_18.up_proj": 2, "layer_19.down_proj": 4, "layer_19.gate_proj": 2, "layer_19.up_proj": 2, "layer_20.down_proj": 4, "layer_20.gate_proj": 2, "layer_20.up_proj": 2, "layer_21.down_proj": 4, "layer_21.gate_proj": 2, "layer_21.up_proj": 2, "layer_22.down_proj": 4, "layer_22.gate_proj": 4, "layer_22.up_proj": 2, "layer_23.down_proj": 4, "layer_23.gate_proj": 4, "layer_23.up_proj": 4, "layer_24.down_proj": 4, "layer_24.gate_proj": 4, "layer_24.up_proj": 4, "layer_25.down_proj": 4, "layer_25.gate_proj": 4, "layer_25.up_proj": 4, "layer_26.down_proj": 4, "layer_26.gate_proj": 4, "layer_26.up_proj": 4, "layer_27.down_proj": 4, "layer_27.gate_proj": 4, "layer_27.up_proj": 4, "layer_28.down_proj": 4, "layer_28.gate_proj": 4, "layer_28.up_proj": 4, "layer_29.down_proj": 4, "layer_29.gate_proj": 4, "layer_29.up_proj": 4, "layer_30.down_proj": 4, "layer_30.gate_proj": 4, "layer_30.up_proj": 4, "layer_31.down_proj": 4, "layer_31.gate_proj": 4, "layer_31.up_proj": 4, "layer_32.down_proj": 4, "layer_32.gate_proj": 4, "layer_32.up_proj": 4, "layer_33.down_proj": 4, "layer_33.gate_proj": 4, "layer_33.up_proj": 4, "layer_34.down_proj": 4, "layer_34.gate_proj": 4, "layer_34.up_proj": 4, "layer_35.down_proj": 4, "layer_35.gate_proj": 4, "layer_35.up_proj": 4, "layer_36.down_proj": 4, "layer_36.gate_proj": 4, "layer_36.up_proj": 4, "layer_37.down_proj": 4, "layer_37.gate_proj": 4, "layer_37.up_proj": 4, "layer_38.down_proj": 4, "layer_38.gate_proj": 4, "layer_38.up_proj": 4, "layer_39.down_proj": 4, "layer_39.gate_proj": 4, "layer_39.up_proj": 4, "layer_40.down_proj": 4, "layer_40.gate_proj": 4, "layer_40.up_proj": 4, "layer_41.down_proj": 4, "layer_41.gate_proj": 4, "layer_41.up_proj": 4, "layer_42.down_proj": 4, "layer_42.gate_proj": 4, "layer_42.up_proj": 4 }, "assignment_unit": "layer.projection, shared across all 256 routed experts" }, "mxtq_bits": { "routed_expert": 2, "attention": 8, "shared_expert": 8, "compressor": 8, "indexer": 8, "embed_tokens": 8, "lm_head": 8, "norms_router_hc": 16 }, "mxtq_seed": 42, "group_size": 32, "_name_or_path": "DSV4F-MLX-TQ-Q3.4" }