[Security] Accept zero-sum DeepSeek-OCR pixel tensors (#59417)
Signed-off-by: Juan Pérez de Algaba <jperezde@redhat.com>
This commit is contained in:
@@ -457,8 +457,6 @@ class DeepseekOCRForCausalLM(
|
||||
return None
|
||||
assert isinstance(pixel_values, torch.Tensor)
|
||||
assert images_crop is None or isinstance(images_crop, torch.Tensor)
|
||||
if torch.sum(pixel_values).item() == 0:
|
||||
return None
|
||||
|
||||
# Use actual tensor spatial dim instead of hardcoded
|
||||
# vision_config.image_size (1024). The vision encoders (SAM & CLIP)
|
||||
|
||||
@@ -318,7 +318,7 @@ class DeepseekOCR2ForCausalLM(nn.Module, SupportsMultiModal, SupportsPP, Support
|
||||
images_spatial_crop = kwargs.pop("images_spatial_crop", None)
|
||||
images_crop = kwargs.pop("images_crop", None)
|
||||
|
||||
if pixel_values is None or torch.sum(pixel_values).item() == 0:
|
||||
if pixel_values is None:
|
||||
return None
|
||||
|
||||
base_size = self.vision_config.image_size
|
||||
|
||||
Reference in New Issue
Block a user