audioldm/clap/open_clip/model_configs/RN50x4.json

{
    "embed_dim": 640,
    "vision_cfg": {
        "image_size": 288,
        "layers": [
            4,
            6,
            10,
            6
        ],
        "width": 80,
        "patch_size": null
    },
    "text_cfg": {
        "context_length": 77,
        "vocab_size": 49408,
        "width": 640,
        "heads": 10,
        "layers": 12
    }
}