[PR #601] [CLOSED] fix(hybrid-ep): replace deprecated pynvml dependency with nvidia-ml-py #654

New issue

Closed

opened 2026-10-27 21:04:56 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-27 21:04:56 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepEP/pull/601
Author: @ooooo-create
Created: 4/8/2026
Status: ❌ Closed

Base: hybrid-ep ← Head: fix/replace-pynvml-with-nvidia-ml-py

📝 Commits (1)

a63558e fix: replace deprecated pynvml dependency with nvidia-ml-py

📊 Changes

1 file changed (+1 additions, -1 deletions)

View changed files

📝 setup.py (+1 -1)

📄 Description

This PR replaces the deprecated pynvml package with nvidia-ml-py>=12.0.0 in setup.py.

pynvml is now deprecated and can trigger a FutureWarning in downstream environments when torch.cuda is imported. Since nvidia-ml-py provides the official NVIDIA-maintained NVML bindings and preserves the pynvml module import path, switching the dependency resolves the warning without changing runtime behavior.

The >=12.0.0 lower bound matches the compatibility floor already implied by the deprecated pynvml package itself.

github.com/gpuopenanalytics/pynvml@493a974f22/pyproject.toml (L15-L17)

_{🔄 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/601 **Author:** [@ooooo-create](https://github.com/ooooo-create) **Created:** 4/8/2026 **Status:** ❌ Closed **Base:** `hybrid-ep` ← **Head:** `fix/replace-pynvml-with-nvidia-ml-py` --- ### 📝 Commits (1) - [`a63558e`](https://github.com/deepseek-ai/DeepEP/commit/a63558e47e2494c55687a3289c0ab49cb0ded940) fix: replace deprecated pynvml dependency with nvidia-ml-py ### 📊 Changes **1 file changed** (+1 additions, -1 deletions) <details> <summary>View changed files</summary> 📝 `setup.py` (+1 -1) </details> ### 📄 Description This PR replaces the deprecated `pynvml` package with `nvidia-ml-py>=12.0.0` in `setup.py`. `pynvml` is now deprecated and can trigger a `FutureWarning` in downstream environments when `torch.cuda` is imported. Since `nvidia-ml-py` provides the official NVIDIA-maintained NVML bindings and preserves the `pynvml` module import path, switching the dependency resolves the warning without changing runtime behavior. The `>=12.0.0` lower bound matches the compatibility floor already implied by the deprecated `pynvml` package itself. - https://github.com/gpuopenanalytics/pynvml/blob/493a974f22ca8c8b30ee0755ae058365cd7ab76c/pyproject.toml#L15-L17 --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-27 21:04:56 +00:00