config.json

{
  "_attn_implementation_autoset": true,
  "_name_or_path": "hf-internal-testing/tiny-random-OwlViTForObjectDetection",
  "architectures": [
    "OwlViTForObjectDetection"
  ],
  "bos_token_id": 1,
  "eos_token_id": 2,
  "initializer_factor": 1.0,
  "logit_scale_init_value": 2.6592,
  "model_type": "owlvit",
  "pad_token_id": 0,
  "projection_dim": 64,
  "text_config": {
    "attention_dropout": 0.1,
    "dropout": 0.1,
    "hidden_size": 64,
    "intermediate_size": 37,
    "model_type": "owlvit_text_model",
    "num_attention_heads": 4,
    "vocab_size": 1024
  },
  "transformers_version": "4.48.2",
  "vision_config": {
    "attention_dropout": 0.1,
    "dropout": 0.1,
    "hidden_size": 32,
    "image_size": 32,
    "intermediate_size": 37,
    "model_type": "owlvit_vision_model",
    "num_attention_heads": 4,
    "num_hidden_layers": 5,
    "patch_size": 2
  }
}