config.json

{
  "model_type": "tiny_clip_t5",
  "vision_config": {
    "hidden_size": 64,
    "num_hidden_layers": 1,
    "num_attention_heads": 8,
    "image_size": 224,
    "patch_size": 16
  },
  "text_config": {
    "vocab_size": 32128,
    "hidden_size": 256,
    "num_hidden_layers": 1,
    "num_attention_heads": 8
  },
  "mm_hidden_size": 256,
  "torch_dtype": "float32"
}