{
  "model_type": "transformer",
  "vocab_size": 32000,
  "num_layers": 12,
  "num_attention_heads": 8,
  "hidden_size": 512,
  "intermediate_size": 2048,
  "max_seq_len": 2048,
  "dropout": 0.1,
  "layer_norm_epsilon": 1e-6
}