Update eva_clip/model.py

eva_clip/model.py

@@ -17,7 +17,7 @@ try:
 except:
     HFTextEncoder = None
 from .modified_resnet import ModifiedResNet
-from .timm_model import TimmModel
+# from .timm_model import TimmModel
 from .eva_vit_model import EVAVisionTransformer
 from .transformer import LayerNorm, QuickGELU, Attention, VisionTransformer, TextTransformer
 
@@ -130,16 +130,17 @@ def _build_vision_tower(
             subln= vision_cfg.subln
         )
     elif vision_cfg.timm_model_name:
-        visual = TimmModel(
-            vision_cfg.timm_model_name,
-            pretrained=vision_cfg.timm_model_pretrained,
-            pool=vision_cfg.timm_pool,
-            proj=vision_cfg.timm_proj,
-            proj_bias=vision_cfg.timm_proj_bias,
-            embed_dim=embed_dim,
-            image_size=vision_cfg.image_size
-        )
-        act_layer = nn.GELU  # so that text transformer doesn't use QuickGELU w/ timm models
+        # visual = TimmModel(
+        #     vision_cfg.timm_model_name,
+        #     pretrained=vision_cfg.timm_model_pretrained,
+        #     pool=vision_cfg.timm_pool,
+        #     proj=vision_cfg.timm_proj,
+        #     proj_bias=vision_cfg.timm_proj_bias,
+        #     embed_dim=embed_dim,
+        #     image_size=vision_cfg.image_size
+        # )
+        # act_layer = nn.GELU  # so that text transformer doesn't use QuickGELU w/ timm models
+        raise ValueError
     elif isinstance(vision_cfg.layers, (tuple, list)):
         vision_heads = vision_cfg.width * 32 // vision_cfg.head_width
         visual = ModifiedResNet(