{
  "model_type": "gpt-neo",
  "architecture": "transformer",
  "hidden_size": 256,
  "num_attention_heads": 8,
  "num_layers": 12,
  "vocab_size": 50257,
  "max_position_embeddings": 1024,
  "type_vocab_size": 2,
  "initializer_range": 0.02,
  "layer_norm_eps": 1e-5,
  "hidden_dropout_prob": 0.1,
  "attention_probs_dropout_prob": 0.1
}