CoolFace
Modelpublic

amd/DeepSeek-R1-MXFP4

sourceHugging Facemitupdated 5mo agoView on Hugging Face
5likes271kdownloads
config.json121 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  "pad_token_id": 128815,36  "pretraining_tp": 1,37  "q_lora_rank": 1536,38  "qk_nope_head_dim": 128,39  "qk_rope_head_dim": 64,40  "quantization_config": {41    "algo_config": null,42    "exclude": [43      "lm_head",44      "re:model.decoder.*",45      "re:model.layers.61.*"46    ],47    "export": {48      "kv_cache_group": [],49      "min_kv_scale": 0.0,50      "pack_method": "reorder",51      "weight_format": "real_quantized",52      "weight_merge_groups": null53    },54    "global_quant_config": {55      "bias": null,56      "input_tensors": {57        "ch_axis": -1,58        "dtype": "fp4",59        "group_size": 32,60        "is_dynamic": true,61        "is_mx_scale_constraint": false,62        "is_scale_quant": false,63        "mx_element_dtype": null,64        "observer_cls": "PerBlockMXObserver",65        "qscheme": "per_group",66        "round_method": "half_even",67        "scale_calculation_mode": "even",68        "scale_format": "e8m0",69        "scale_type": "float",70        "symmetric": null71      },72      "output_tensors": null,73      "target_device": null,74      "weight": {75        "ch_axis": -1,76        "dtype": "fp4",77        "group_size": 32,78        "is_dynamic": false,79        "is_mx_scale_constraint": false,80        "is_scale_quant": false,81        "mx_element_dtype": null,82        "observer_cls": "PerBlockMXObserver",83        "qscheme": "per_group",84        "round_method": "half_even",85        "scale_calculation_mode": "even",86        "scale_format": "e8m0",87        "scale_type": "float",88        "symmetric": null89      }90    },91    "layer_quant_config": {},92    "layer_type_quant_config": {},93    "quant_method": "quark",94    "quant_mode": "eager_mode",95    "softmax_quant_spec": null96  },97  "rms_norm_eps": 1e-06,98  "rope_scaling": {99    "beta_fast": 32,100    "beta_slow": 1,101    "factor": 40,102    "mscale": 1.0,103    "mscale_all_dim": 1.0,104    "original_max_position_embeddings": 4096,105    "type": "yarn"106  },107  "rope_theta": 10000,108  "routed_scaling_factor": 2.5,109  "scoring_func": "sigmoid",110  "seq_aux": true,111  "tie_word_embeddings": false,112  "topk_group": 4,113  "topk_method": "noaux_tc",114  "torch_dtype": "bfloat16",115  "transformers_version": "4.51.0",116  "unsloth_fixed": true,117  "use_cache": true,118  "v_head_dim": 128,119  "vocab_size": 129280120}121