config.json

{
  "architectures": [
    "PeVideoEncoder"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "dtype": "float32",
  "head_dim": 16,
  "hidden_act": "silu",
  "hidden_size": 32,
  "initializer_range": 0.02,
  "intermediate_size": 37,
  "max_position_embeddings": 512,
  "max_window_layers": 28,
  "model_type": "pe_video_encoder",
  "num_attention_heads": 2,
  "num_hidden_layers": 2,
  "num_key_value_heads": 2,
  "rms_norm_eps": 1e-05,
  "rope_parameters": {
    "rope_theta": 20000,
    "rope_type": "default"
  },
  "transformers_version": "5.16.0.dev0",
  "vision_config": {
    "architecture": "vit_pe_core_large_patch14_336",
    "do_pooling": true,
    "global_pool": "map",
    "initializer_range": 0.02,
    "label_names": [
      "LABEL_0",
      "LABEL_1",
      "LABEL_2",
      "LABEL_3"
    ],
    "model_args": {
      "depth": 2,
      "embed_dim": 64,
      "img_size": [
        14,
        14
      ]
    },
    "model_type": "timm_wrapper",
    "num_classes": 4
  }
}