Skip to content

fix multinode dp mode bug. #986

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

Merged
hiworldwzj merged 1 commit into from
Jul 24, 2025
Merged

fix multinode dp mode bug. #986

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
4 changes: 3 additions & 1 deletion lightllm/server/router/model_infer/mode_backend/base_backend.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -282,7 +282,9 @@ def _try_read_new_reqs_normal(self):
self.node_broadcast_tensor.fill_(1)
else:
self.node_broadcast_tensor.fill_(0)
dist.broadcast(self.node_broadcast_tensor, src=0, group=self.node_nccl_group, async_op=False)

src_rank_id = self.args.node_rank * self.node_world_size
Copy link
Contributor

@gemini-code-assist gemini-code-assist bot Jul 24, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

The calculation for src_rank_id can be simplified by deriving it from the existing rank properties of the class. The global rank of the master process in the node can be calculated as self.global_rank - self.rank_in_node. This avoids a direct dependency on self.args and leverages the established distributed rank properties, which can improve code consistency and maintainability.

src_rank_id = self.global_rank - self.rank_in_node

dist.broadcast(self.node_broadcast_tensor, src=src_rank_id, group=self.node_nccl_group, async_op=False)
new_buffer_is_ready = self.node_broadcast_tensor.detach().item()
if new_buffer_is_ready:
self._read_reqs_buffer_and_init_reqs()
Expand Down