config.json

{
  "apply_l2norm": true,
  "architectures": [
    "VideoPrismForVideoClassification"
  ],
  "attention_probs_dropout_prob": 0.0,
  "attn_logit_softcapping": 50.0,
  "dtype": "float32",
  "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,
  "transformers_version": "5.16.0.dev0",
  "tubelet_size": [
    1,
    4,
    4
  ]
}