mixvideo-v2/cargos/tvai-v2/视觉-语言模型配置/ViT-B-16-plus-240.json

{
    "embed_dim": 640,
    "vision_cfg": {
        "image_size": 240,
        "layers": 12,
        "width": 896,
        "patch_size": 16
    },
    "text_cfg": {
        "context_length": 77,
        "vocab_size": 49408,
        "width": 640,
        "heads": 10,
        "layers": 12
    }
}