config.json

{
  "architectures": [
    "DeepSeekMoE"
  ],
  "block_size": 2048,
  "dtype": "bfloat16",
  "expert_intermediate_size": 1408,
  "head_dim": 64,
  "model_type": "eve_moe",
  "n_embd": 512,
  "n_head": 8,
  "n_layer": 12,
  "num_experts": 8,
  "rope_theta": 10000.0,
  "router_aux_loss_coef": 0.01,
  "shared_expert_intermediate_size": 1408,
  "top_k": 2,
  "transformers_version": "5.1.0",
  "use_cache": false,
  "use_checkpointing": false,
  "vocab_size": 50304
}