config.json

{
  "architectures": [
    "VideoPrismClipModel"
  ],
  "dtype": "float32",
  "model_type": "videoprism",
  "text_config": {
    "apply_l2norm": true,
    "attention_probs_dropout_prob": 0.0,
    "attn_logit_softcapping": 50.0,
    "bos_token_id": null,
    "eos_token_id": null,
    "hidden_act": "relu",
    "hidden_dropout_prob": 0.0,
    "hidden_size": 32,
    "initializer_range": 0.02,
    "intermediate_size": 37,
    "layer_norm_eps": 1e-06,
    "max_position_embeddings": 200,
    "model_type": "videoprism_text_model",
    "num_attention_heads": 4,
    "num_hidden_layers": 2,
    "pad_token_id": 0,
    "qkv_bias": true,
    "vocab_size": 32101
  },
  "transformers_version": "5.16.0.dev0",
  "vision_config": {
    "apply_l2norm": true,
    "attention_probs_dropout_prob": 0.0,
    "attn_logit_softcapping": 50.0,
    "hidden_act": "gelu_python",
    "hidden_dropout_prob": 0.0,
    "hidden_size": 32,
    "image_size": 8,
    "initializer_range": 0.02,
    "intermediate_size": 64,
    "layer_norm_eps": 1e-06,
    "model_type": "videoprism_vision_model",
    "num_attention_heads": 4,
    "num_auxiliary_layers": 2,
    "num_channels": 3,
    "num_frames": 3,
    "num_spatial_layers": 3,
    "num_temporal_layers": 2,
    "qkv_bias": true,
    "tubelet_size": [
      1,
      4,
      4
    ]
  }
}