Skip to content

fix(dp): wait forward before gather_req #1506

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

Open
WANDY666 wants to merge 1 commit into
base: main
Choose a base branch
from
+6 −0
Open

fix(dp): wait forward before gather_req #1506

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
6 changes: 6 additions & 0 deletions lightllm/server/router/model_infer/mode_backend/dp_backend/impl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -130,6 +130,12 @@ def infer_loop(self):
recover_paused=self.control_state_machine.try_recover_paused_reqs(),
)

if self.support_overlap:
# The previous infer thread releases forward before its GPU work finishes.
# Keep the next DP collective behind that work to avoid overlapping NCCL
# with the previous DeepEP/MTP kernels.
torch.cuda.current_stream().wait_stream(g_infer_context.get_overlap_stream())

dp_prefill_req_nums, dp_decode_req_nums = self._dp_all_gather_prefill_and_decode_req_num(
prefill_reqs=prefill_reqs, decode_reqs=decode_reqs
)
Expand Down
Loading