Skip to content

[Bugfix] Fix race condition in MTP speculative decoding draft loop #42603

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
z1ying wants to merge 2 commits into from
Closed

[Bugfix] Fix race condition in MTP speculative decoding draft loop #42603

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9d7ca1d
add current stream sync to prevent race condition
z1ying May 14, 2026
cccbad3
Merge branch 'main' into Bugfix/CUDA-stream-race-condition-in-MTP-seq…
z1ying May 14, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 7 additions & 0 deletions vllm/v1/spec_decode/llm_base_proposer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -589,6 +589,13 @@ def propose(
# copy inputs to buffer for cudagraph
self.input_ids[:batch_size] = input_ids
self.hidden_states[:batch_size] = hidden_states

# Wait for the `input_ids` write to finish on device before constructing
# attention metadata or launching downstream kernels. Synchronizing the
# current stream blocks the CPU minimally, avoiding an expensive
# global torch.accelerator.synchronize().
torch.accelerator.current_stream().synchronize()

if self.supports_mm_inputs:
self.inputs_embeds[:batch_size] = self.model.embed_input_ids(input_ids)

Expand Down
Loading