distil-test / config.json
Godwinlyamba's picture
Upload folder using huggingface_hub
673da73 verified
{
"transformers_version": "5.7.0",
"architectures": [
"DeepseekV3ForCausalLM"
],
"output_hidden_states": false,
"return_dict": true,
"dtype": "bfloat16",
"chunk_size_feed_forward": 0,
"is_encoder_decoder": false,
"id2label": {
"0": "LABEL_0",
"1": "LABEL_1"
},
"label2id": {
"LABEL_0": 0,
"LABEL_1": 1
},
"problem_type": null,
"vocab_size": 163840,
"hidden_size": 2048,
"intermediate_size": 11264,
"moe_intermediate_size": 1408,
"num_hidden_layers": 27,
"num_attention_heads": 16,
"num_key_value_heads": 16,
"n_shared_experts": 2,
"n_routed_experts": 64,
"routed_scaling_factor": 2.446,
"kv_lora_rank": 512,
"q_lora_rank": null,
"qk_rope_head_dim": 64,
"v_head_dim": 128,
"qk_nope_head_dim": 128,
"n_group": 1,
"topk_group": 1,
"num_experts_per_tok": 6,
"first_k_dense_replace": 1,
"norm_topk_prob": true,
"hidden_act": "silu",
"max_position_embeddings": 131072,
"initializer_range": 0.02,
"rms_norm_eps": 1e-05,
"use_cache": false,
"pad_token_id": 163839,
"bos_token_id": 163584,
"eos_token_id": 163585,
"pretraining_tp": 1,
"tie_word_embeddings": false,
"rope_parameters": {
"rope_theta": 800000.0,
"rope_type": "default"
},
"rope_interleave": true,
"attention_bias": false,
"attention_dropout": 0.0,
"qk_head_dim": 192,
"head_dim": 64,
"_name_or_path": "/ephemeral/ubuntu/distil-metadata/model_b",
"aux_loss_alpha": 0.001,
"ep_size": 1,
"model_type": "deepseek_v3",
"moe_layer_freq": 1,
"num_nextn_predict_layers": 1,
"scoring_func": "sigmoid",
"seq_aux": true,
"topk_method": "noaux_tc",
"output_attentions": false,
"torch_dtype": "bfloat16"
}