config.json

{
    "audio_encoder": {
        "embed_dim_audio": 1024,
        "joint_dim": 512,
        "audio_cfg": {
            "audio_length": 1024,
            "clip_samples": 480000,
            "mel_bins": 64,
            "sample_rate": 48000,
            "window_size": 1024,
            "hop_size": 480,
            "fmin": 50,
            "fmax": 14000,
            "class_num": 527,
            "model_type": "HTSAT",
            "model_name": "base"
        }
    },
    "LLM": {
        "t5_model": "google/flan-t5-xl",
        "repetition_penalty": 10.0
    },
    "num_query_tokens": 16,
    "embed_dim": 256,
    "max_txt_len": 120,
    "prompt": "Caption this data:",
    "apply_lemmatizer": false,
    "atRandom": true
}