|
|
@@ -6145,7 +6145,8 @@ class Gemma3nVisionAudioModel(ConformerAudioModel):
|
|
|
|
|
|
if name.startswith("model.vision_tower.timm_model.blocks."):
|
|
|
# Double-indexed block tensors through custom logic
|
|
|
- new_name = self.custom_map(name)
|
|
|
+ yield (self.custom_map(name), data_torch)
|
|
|
+ return
|
|
|
else:
|
|
|
# Route non-repeating (conv_stem, msfa, embedding, etc.) and un-catched through tensor_mapping.py
|
|
|
new_name = self.map_tensor_name(name)
|