config.json

{
  "architectures": [
    "SentinelBrainForCausalLM"
  ],
  "model_type": "sentinelbrain",
  "vocab_size": 100277,
  "hidden_size": 4096,
  "num_hidden_layers": 24,
  "num_attention_heads": 32,
  "num_key_value_heads": 8,
  "intermediate_size": 11008,
  "max_position_embeddings": 2048,
  "rope_theta": 500000.0,
  "rms_norm_eps": 1e-05,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "num_local_experts": 4,
  "num_experts_per_tok": 2,
  "max_experts": 256,
  "routing_mode": "token_choice",
  "expert_capacity_factor": 1.25,
  "router_aux_loss_coeff": 0.01,
  "router_z_loss_coeff": 0.001,
  "training_step": 2471,
  "training_val_loss": 1.9925728058815002,
  "training_tokens": 178110464
}