Upload OrthoViTForImageClassification

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a0d99e8aebb9719333f101b00f1110d881d3e084c80afbf8ca96c45891449b0a
+oid sha256:9a9735341b0c6e7e502b992b918dee50943a643182913e280e5d0308beda20e9
 size 346141064

modeling_ortho_vit.py

@@ -379,12 +379,11 @@ class OrthoViTForImageClassification(OrthoViTPreTrainedModel):
         self.vit = OrthoViTModel(config, add_pooling_layer=False) # Use OrthoViTModel
 
         # Classifier head (standard HF: just a Linear layer)
-        # The user's original classifier was nn.Sequential(nn.LayerNorm(dim), nn.Linear(dim, num_classes))
+        # The author's original classifier was nn.Sequential(nn.LayerNorm(dim), nn.Linear(dim, num_classes))
         # The OrthoViTModel's output (CLS token) is already layernormed by self.vit.layernorm.
-        # So, if that self.vit.layernorm corresponds to the user's classifier.0 (LayerNorm),
+        # So, if that self.vit.layernorm corresponds to the author's classifier.0 (LayerNorm),
         # then this classifier head should just be Linear.
         # If an *additional* LN is needed on the CLS token, this needs to be nn.Sequential.
-        # For now, assume vit.layernorm covers the user's classifier.0.
         self.classifier = nn.Linear(config.hidden_size, config.num_labels) if config.num_labels > 0 else nn.Identity()
 
         # Initialize weights and apply final processing