{
  "architectures": ["FleckModel"],
  "model_type": "fleck",
  "custom_architecture": true,
  "implementation": "inference.py",
  "candidate_id": "Fleck-S-100K-Base",
  "vocab_size": 1024,
  "hidden_size": 64,
  "intermediate_size": 128,
  "embedding_rank": 32,
  "num_hidden_layers": 4,
  "num_physical_blocks": 2,
  "num_attention_heads": 4,
  "num_key_value_heads": 2,
  "head_dim": 16,
  "attention_type": "GQA",
  "execution_order": ["A", "B", "A", "B"],
  "embedding_mode": "factorized_tied",
  "tie_word_embeddings": true,
  "max_position_embeddings": 2048,
  "torch_dtype": "bfloat16",
  "bos_token_id": 0,
  "eos_token_id": 1,
  "pad_token_id": 2,
  "unk_token_id": 3,
  "use_cache": true
}