hf/config.json

{
  "architectures": [
    "DeepseekV2ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "aux_loss_alpha": 0.01,
  "bos_token_id": 1000,
  "dtype": "float32",
  "eos_token_id": 1001,
  "first_k_dense_replace": 1,
  "head_dim": 16,
  "hidden_act": "silu",
  "hidden_size": 216,
  "initializer_range": 0.02,
  "intermediate_size": 432,
  "kv_lora_rank": 64,
  "max_position_embeddings": 2048,
  "mlp_bias": false,
  "model_type": "deepseek_v2",
  "moe_intermediate_size": 128,
  "n_group": null,
  "n_routed_experts": 4,
  "n_shared_experts": 1,
  "norm_topk_prob": false,
  "num_attention_heads": 8,
  "num_experts_per_tok": 1,
  "num_hidden_layers": 5,
  "num_key_value_heads": 8,
  "pad_token_id": 1000,
  "pretraining_tp": 1,
  "q_lora_rank": null,
  "qk_nope_head_dim": 16,
  "qk_rope_head_dim": 16,
  "rms_norm_eps": 1e-06,
  "rope_parameters": {
    "rope_theta": 10000,
    "rope_type": "default"
  },
  "routed_scaling_factor": 4.0,
  "seq_aux": true,
  "tie_word_embeddings": true,
  "topk_group": null,
  "topk_method": "greedy",
  "transformers_version": "5.14.1",
  "use_cache": false,
  "v_head_dim": 32,
  "vocab_size": 1024
}