config.json

{
  "ablate": [
    "none"
  ],
  "architectures": [
    "MiCRoLlamaMoE"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "backbone_num_layers": 16,
  "bos_token_id": 128000,
  "config_path": "../mixture-of-reasoners/configs/config_micro_moe_llama.yml",
  "eos_token_id": 128001,
  "gradient_checkpointing": false,
  "head_dim": 64,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 8192,
  "jitter_noise": 0,
  "loss_method": "all",
  "loss_type": "ForCausalLMLoss",
  "max_position_embeddings": 131072,
  "mlp_bias": false,
  "model_type": "llama",
  "num_attention_heads": 32,
  "num_experts": 4,
  "num_experts_per_tok": 1,
  "num_hidden_layers": 16,
  "num_key_value_heads": 8,
  "pretraining_tp": 1,
  "rms_norm_eps": 1e-05,
  "rope_scaling": {
    "factor": 32.0,
    "high_freq_factor": 4.0,
    "low_freq_factor": 1.0,
    "original_max_position_embeddings": 8192,
    "rope_type": "llama3"
  },
  "rope_theta": 500000.0,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.53.2",
  "use_bfloat16": true,
  "use_cache": true,
  "use_router": true,
  "vocab_size": 128256
}