Add vit_little in12k + in12k-ft-in1k weights

huggingface · May 27, 2024 · 5dce710 · 5dce710
1 parent 3c0283f
commit 5dce710
Showing 1 changed file with 18 additions and 0 deletions.
diff --git a/timm/models/vision_transformer.py b/timm/models/vision_transformer.py
@@ -1872,6 +1872,13 @@ def _cfg(url: str = '', **kwargs) -> Dict[str, Any]:
     'vit_pwee_patch16_reg1_gap_256.sbb_in1k': _cfg(
         hf_hub_id='timm/',
         input_size=(3, 256, 256), crop_pct=0.95),
+    'vit_little_patch16_reg1_gap_256.sbb_in12k_ft_in1k': _cfg(
+        hf_hub_id='timm/',
+        input_size=(3, 256, 256), crop_pct=0.95),
+    'vit_little_patch16_reg1_gap_256.sbb_in12k': _cfg(
+        hf_hub_id='timm/',
+        num_classes=11821,
+        input_size=(3, 256, 256), crop_pct=0.95),
     'vit_little_patch16_reg4_gap_256.sbb_in1k': _cfg(
         hf_hub_id='timm/',
         input_size=(3, 256, 256), crop_pct=0.95),
@@ -2975,6 +2982,17 @@ def vit_pwee_patch16_reg1_gap_256(pretrained: bool = False, **kwargs) -> VisionT
     return model
 
 
+@register_model
+def vit_little_patch16_reg1_gap_256(pretrained: bool = False, **kwargs) -> VisionTransformer:
+    model_args = dict(
+        patch_size=16, embed_dim=320, depth=14, num_heads=5, init_values=1e-5, mlp_ratio=5.6,
+        class_token=False, no_embed_class=True, reg_tokens=1, global_pool='avg',
+    )
+    model = _create_vision_transformer(
+        'vit_little_patch16_reg1_gap_256', pretrained=pretrained, **dict(model_args, **kwargs))
+    return model
+
+
 @register_model
 def vit_little_patch16_reg4_gap_256(pretrained: bool = False, **kwargs) -> VisionTransformer:
     model_args = dict(