final_model/config.json

{
  "activation_function": "gelu",
  "architectures": [
    "MoEGPTForCausalLM"
  ],
  "aux_loss_alpha": 0.01,
  "bias": false,
  "block_size": 2048,
  "bos_token_id": 128000,
  "capacity_factor": 1.25,
  "dropout": 0.0,
  "dtype": "float32",
  "eos_token_id": 128001,
  "eval_capacity_factor": 2.0,
  "hidden_size": 768,
  "initializer_range": 0.1,
  "intermediate_size": 2048,
  "layer_norm_epsilon": 1e-05,
  "max_position_embeddings": 2048,
  "model_type": "moe_gpt",
  "moe_layer_frequency": 2,
  "n_embd": 768,
  "n_experts": 8,
  "n_experts_active": 2,
  "n_head": 12,
  "n_layer": 12,
  "n_positions": 2048,
  "num_attention_heads": 12,
  "num_hidden_layers": 12,
  "pad_token_id": 128001,
  "rope_theta": 10000.0,
  "router_z_loss_alpha": 0.001,
  "transformers_version": "4.57.1",
  "use_cache": true,
  "use_noisy_gating": true,
  "use_rope": true,
  "vocab_size": 128256
}