[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
@@ -61,6 +61,7 @@ from vllm.multimodal.processing import (
|
||||
TimingContext,
|
||||
)
|
||||
from vllm.sequence import IntermediateTensors
|
||||
from vllm.utils.gpu_sync_debug import gpu_sync_allowed
|
||||
|
||||
from .interfaces import IsAttentionFree, MultiModalEmbeddings, SupportsMultiModal
|
||||
from .interfaces_base import attn_type
|
||||
@@ -280,7 +281,9 @@ class Terratorch(nn.Module, IsAttentionFree, SupportsMultiModal):
|
||||
inputs_embeds: torch.Tensor | None = None,
|
||||
**kwargs: object,
|
||||
):
|
||||
model_output = self.inference_runner.forward(**kwargs)
|
||||
# terratorch's forward has internal GPU syncs.
|
||||
with gpu_sync_allowed():
|
||||
model_output = self.inference_runner.forward(**kwargs)
|
||||
return model_output.output
|
||||
|
||||
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
|
||||
|
||||
Reference in New Issue
Block a user