inference-optimization/DeepSeek-V3-debug-empty

1{2  "architectures": [3    "DeepseekV3ForCausalLM"4  ],5  "attention_bias": false,6  "attention_dropout": 0.0,7  "bos_token_id": 0,8  "dtype": "bfloat16",9  "eos_token_id": 1,10  "ep_size": 1,11  "first_k_dense_replace": 0,12  "head_dim": 64,13  "hidden_act": "silu",14  "hidden_size": 1792,15  "initializer_range": 0.02,16  "intermediate_size": 2304,17  "kv_lora_rank": 512,18  "max_position_embeddings": 163840,19  "model_type": "deepseek_v3",20  "moe_intermediate_size": 512,21  "moe_layer_freq": 1,22  "n_group": 8,23  "n_routed_experts": 256,24  "n_shared_experts": 1,25  "norm_topk_prob": true,26  "num_attention_heads": 128,27  "num_experts_per_tok": 8,28  "num_hidden_layers": 1,29  "num_key_value_heads": 128,30  "num_nextn_predict_layers": 1,31  "pretraining_tp": 1,32  "q_lora_rank": 1536,33  "qk_head_dim": 192,34  "qk_nope_head_dim": 128,35  "qk_rope_head_dim": 64,36  "rms_norm_eps": 1e-06,37  "rope_interleave": true,38  "rope_scaling": {39    "beta_fast": 32.0,40    "beta_slow": 1.0,41    "factor": 40.0,42    "mscale": 1.0,43    "mscale_all_dim": 1.0,44    "original_max_position_embeddings": 4096,45    "rope_type": "yarn",46    "type": "yarn"47  },48  "rope_theta": 10000,49  "routed_scaling_factor": 2.5,50  "scoring_func": "sigmoid",51  "tie_word_embeddings": false,52  "topk_group": 4,53  "topk_method": "noaux_tc",54  "transformers_version": "4.57.6",55  "use_cache": true,56  "v_head_dim": 128,57  "vocab_size": 12800058}59