utils/model_configs/vit_medium_patch16_gap_256.json

{
    "embed_dim": 512,
    "vision_cfg": {
        "timm_model_name": "vit_medium_patch16_gap_256",
        "timm_model_pretrained": false,
        "timm_pool": "",
        "timm_proj": "linear",
        "image_size": 256
    },
    "text_cfg": {
        "context_length": 77,
        "vocab_size": 49408,
        "width": 512,
        "heads": 8,
        "layers": 12
    }
}