config.json

{
  "model_type": "microtrillium",
  "architectures": [
    "MicroTrilliumThinkerForCausalLM"
  ],
  "vocab_size": 49152,
  "hidden_size": 1280,
  "num_hidden_layers": 10,
  "num_key_value_heads": 5,
  "num_attention_heads": 20,
  "head_dim": 64,
  "intermediate_size": 3584,
  "max_position_embeddings": 2048,
  "rms_norm_eps": 1e-06,
  "tie_word_embeddings": false,
  "attention_type": "diff_gqa",
  "num_loops": 3,
  "mtp_heads": 2,
  "mtp_weight": 0.3,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.40.0"
}