config.json

{
  "architectures": [
    "ArlowVLVisionModel"
  ],
  "deepstack_visual_indexes": [
    12,
    24,
    44
  ],
  "deformable_attention_strength": 4.0,
  "deformable_attention_window": 0.25,
  "depth": 48,
  "dtype": "float16",
  "embed_dim": 1536,
  "hidden_act": "gelu_pytorch_tanh",
  "hidden_size": 3072,
  "in_channels": 3,
  "initializer_range": 0.02,
  "max_position_embeddings": 32768,
  "mlp_ratio": 4,
  "model_type": "arlow",
  "mrope_sections": [
    21,
    21,
    22
  ],
  "num_attention_heads": 24,
  "num_heads": 24,
  "patch_size": 14,
  "spatial_merge_size": 2,
  "temporal_patch_size": 2,
  "token_pruning_ratio": 0.0,
  "transformers_version": "5.3.0.dev0",
  "use_deformable_attention": true,
  "use_progressive_patches": true
}