CoolFace
Modelpublic

meituan/DeepSeek-R1-Block-INT8

sourceHugging Facemitupdated 2y agoView on Hugging Face
50likes866downloads
config.json70 linesDownload Raw Back to root
1{2    "architectures": [3      "DeepseekV3ForCausalLM"4    ],5    "attention_bias": false,6    "attention_dropout": 0.0,7    "auto_map": {8      "AutoConfig": "configuration_deepseek.DeepseekV3Config",9      "AutoModel": "modeling_deepseek.DeepseekV3Model",10      "AutoModelForCausalLM": "modeling_deepseek.DeepseekV3ForCausalLM"11    },12    "aux_loss_alpha": 0.001,13    "bos_token_id": 0,14    "eos_token_id": 1,15    "ep_size": 1,16    "first_k_dense_replace": 3,17    "hidden_act": "silu",18    "hidden_size": 7168,19    "initializer_range": 0.02,20    "intermediate_size": 18432,21    "kv_lora_rank": 512,22    "max_position_embeddings": 163840,23    "model_type": "deepseek_v3",24    "moe_intermediate_size": 2048,25    "moe_layer_freq": 1,26    "n_group": 8,27    "n_routed_experts": 256,28    "n_shared_experts": 1,29    "norm_topk_prob": true,30    "num_attention_heads": 128,31    "num_experts_per_tok": 8,32    "num_hidden_layers": 61,33    "num_key_value_heads": 128,34    "num_nextn_predict_layers": 1,35    "pretraining_tp": 1,36    "q_lora_rank": 1536,37    "qk_nope_head_dim": 128,38    "qk_rope_head_dim": 64,39    "quantization_config": {40      "activation_scheme": "dynamic",41      "quant_method": "blockwise_int8",42      "weight_block_size": [43        128,44        12845      ]46    },47    "rms_norm_eps": 1e-06,48    "rope_scaling": {49      "beta_fast": 32,50      "beta_slow": 1,51      "factor": 40,52      "mscale": 1.0,53      "mscale_all_dim": 1.0,54      "original_max_position_embeddings": 4096,55      "type": "yarn"56    },57    "rope_theta": 10000,58    "routed_scaling_factor": 2.5,59    "scoring_func": "sigmoid",60    "seq_aux": true,61    "tie_word_embeddings": false,62    "topk_group": 4,63    "topk_method": "noaux_tc",64    "torch_dtype": "bfloat16",65    "transformers_version": "4.46.3",66    "use_cache": true,67    "v_head_dim": 128,68    "vocab_size": 12928069  }70