[PR #646] make ginQueueDepth configurable #686

New issue

Open

opened 2026-10-27 21:05:16 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-27 21:05:16 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepEP/pull/646
Author: @GuangguanWang
Created: 5/26/2026
Status: 🔄 Open

Base: main ← Head: upstream

📝 Commits (2)

18525a6 make ginQueueDepth configurable as GIN PROXY mode requires ginQueueDepth = 0
44941a7 fix for code format

📊 Changes

2 files changed (+30 additions, -30 deletions)

View changed files

📝 README.md (+1 -0)
📝 csrc/kernels/backend/nccl.cu (+29 -30)

📄 Description

make ginQueueDepth configurable as GIN PROXY mode requires ginQueueDepth = 0

relative error msg when run in GIN PROXY mode with 1024 ginQueueDepth

Traceback (most recent call last):
  File "/root/miniconda3/envs/deepepv2/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 87, in _wrap
    fn(i, *args)
  File "/root/miniconda3/envs/deepepv2/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 124, in decorate_context
    return func(*args, **kwargs)
  File "/mnt/deepep-v2/DeepEP/tests/elastic/test_barrier.py", line 36, in test_loop
    buffer = deep_ep.ElasticBuffer(
  File "/root/miniconda3/envs/deepepv2/lib/python3.10/site-packages/deep_ep/buffers/elastic.py", line 217, in __init__
    self.runtime = _C.ElasticBuffer(group.rank(), group.size(),
RuntimeError: NCCL exception (/mnt/deepep-v2/DeepEP/csrc/kernels/backend/nccl.cu:105): 5 (GIN_IB_PROXY does not support specifying qp depth)

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/deepseek-ai/DeepEP/pull/646 **Author:** [@GuangguanWang](https://github.com/GuangguanWang) **Created:** 5/26/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `upstream` --- ### 📝 Commits (2) - [`18525a6`](https://github.com/deepseek-ai/DeepEP/commit/18525a64c9ef77050503719c1a06298b32401aad) make ginQueueDepth configurable as GIN PROXY mode requires ginQueueDepth = 0 - [`44941a7`](https://github.com/deepseek-ai/DeepEP/commit/44941a76e9600d7096087ae5378a44056f0e20b0) fix for code format ### 📊 Changes **2 files changed** (+30 additions, -30 deletions) <details> <summary>View changed files</summary> 📝 `README.md` (+1 -0) 📝 `csrc/kernels/backend/nccl.cu` (+29 -30) </details> ### 📄 Description make ginQueueDepth configurable as GIN PROXY mode requires ginQueueDepth = 0 relative error msg when run in GIN PROXY mode with 1024 ginQueueDepth ``` Traceback (most recent call last): File "/root/miniconda3/envs/deepepv2/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 87, in _wrap fn(i, *args) File "/root/miniconda3/envs/deepepv2/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 124, in decorate_context return func(*args, **kwargs) File "/mnt/deepep-v2/DeepEP/tests/elastic/test_barrier.py", line 36, in test_loop buffer = deep_ep.ElasticBuffer( File "/root/miniconda3/envs/deepepv2/lib/python3.10/site-packages/deep_ep/buffers/elastic.py", line 217, in __init__ self.runtime = _C.ElasticBuffer(group.rank(), group.size(), RuntimeError: NCCL exception (/mnt/deepep-v2/DeepEP/csrc/kernels/backend/nccl.cu:105): 5 (GIN_IB_PROXY does not support specifying qp depth) ``` --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi added the

pull-request

label

2026-10-27 21:05:16 +00:00

bsakaguchi referenced this issue

2026-10-27 21:05:28 +00:00

[PR #686] Compatible with cu132 #708