[Core] Check for GPU<->CPU syncs during CI (#43107)
Signed-off-by: Nick Hill <nickhill123@gmail.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Opus 4.8
parent
7b544ecb52
commit
615d4cfade
@@ -51,6 +51,7 @@ from vllm.multimodal.processing import (
|
||||
PromptUpdateDetails,
|
||||
)
|
||||
from vllm.sequence import IntermediateTensors
|
||||
from vllm.utils.gpu_sync_debug import gpu_sync_allowed
|
||||
from vllm.utils.tensor_schema import TensorSchema, TensorShape
|
||||
|
||||
from .idefics2_vision_model import (
|
||||
@@ -443,11 +444,12 @@ class Idefics3Model(nn.Module):
|
||||
real_images_inds = (pixel_values == 0.0).sum(
|
||||
dim=(-1, -2, -3)
|
||||
) != nb_values_per_image
|
||||
pixel_values = pixel_values[real_images_inds].contiguous()
|
||||
with gpu_sync_allowed():
|
||||
pixel_values = pixel_values[real_images_inds].contiguous()
|
||||
|
||||
# Handle the vision attention mask
|
||||
# Remove padding images from the mask
|
||||
pixel_attention_mask = pixel_attention_mask[real_images_inds].contiguous()
|
||||
# Handle the vision attention mask
|
||||
# Remove padding images from the mask
|
||||
pixel_attention_mask = pixel_attention_mask[real_images_inds].contiguous()
|
||||
|
||||
patch_size = self.config.vision_config.patch_size
|
||||
patches_subgrid = pixel_attention_mask.unfold(
|
||||
|
||||
Reference in New Issue
Block a user