{
    "architecture": "WhisperEncoder",
    "dtype": "float16",
    "num_hidden_layers": 32,
    "num_attention_heads": 20,
    "hidden_size": 1280,
    "n_mels": 128,
    "n_audio_ctx": 1500,
    "vocab_size": 51866,
    "hidden_act": "gelu",
    "num_languages": 100,
    "quantization": {
        "quant_algo": "W8A16"
    }
}