deepseek-ai / deepseek-ai/DeepEP

DeepEP v1 Normal mode, run test_internode RuntimeError: DeepEP error: timeout (dispatch CPU) when use NVSHMEM version >= 3.5.19

Open
#707 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Cuda
Stars
10.1k
Forks
1.4k
Avg merge
2d 17h
Merged PRs (30d)
3

Description

https://github.com/deepseek-ai/DeepEP/blob/dd758caf451848bd150e1046af3d0a73e5fff38d/csrc/kernels/legacy/ibgda_device.cuh#L81

The main branch is missing a previous adaptation for nvshmem 3.5.19 and later versions.
PR: https://github.com/deepseek-ai/DeepEP/pull/564/changes

This causes the current main branch to hang directly in the `notify_dispatch` kernel when running DeepEP v1.

Contributor guide

No contributing guide indexed for this repository

Research direction

Start with csrc/kernels/legacy/ibgda_device.cuh at line 81 and the notify_dispatch kernel, then compare the current main branch with the adaptation referenced in PR #564. Reproduce test_internode with NVSHMEM 3.5.19 or later; done means DeepEP v1 Normal mode no longer hangs or reports a dispatch CPU timeout.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
distributed-systems
Issue type
Bug
Difficulty
3/5
Estimated time
1-2 days
Activity status
Stale
Clarity
Clearly specified
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.