[PR #534] [CLOSED] Add DEEPEP_SKIP_KINETO env var to bypass kineto for nsys profiling #618

New issue

Closed

opened 2026-10-27 21:04:35 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-27 21:04:35 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepEP/pull/534
Author: @yurekami
Created: 12/25/2025
Status: ❌ Closed

Base: main ← Head: feature/skip-kineto-for-nsys

📝 Commits (1)

633fe18 Add DEEPEP_SKIP_KINETO env var to bypass kineto profiling

📊 Changes

1 file changed (+22 additions, -0 deletions)

View changed files

📝 tests/utils.py (+22 -0)

📄 Description

Summary

Add DEEPEP_SKIP_KINETO environment variable to bypass kineto profiling
When set to 1, uses simple CUDA event timing instead of torch.profiler
Allows using external profilers like nsys profile -t cuda without conflicts

Usage

DEEPEP_SKIP_KINETO=1 nsys profile -t cuda python tests/test_internode.py

Background

The test scripts use kineto (torch.profiler) for capturing kernel durations. This can conflict with nsys profile when users want to analyze the kernels with NVIDIA Nsight Systems.

With this change, users can set DEEPEP_SKIP_KINETO=1 to disable kineto and use nsys freely.

Test plan

Run tests with DEEPEP_SKIP_KINETO=1 and verify they complete
Run nsys profile -t cuda with this env var set

Addresses #516

🤖 Generated with Claude Code

_{🔄 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/534 **Author:** [@yurekami](https://github.com/yurekami) **Created:** 12/25/2025 **Status:** ❌ Closed **Base:** `main` ← **Head:** `feature/skip-kineto-for-nsys` --- ### 📝 Commits (1) - [`633fe18`](https://github.com/deepseek-ai/DeepEP/commit/633fe1816edcb0fdc5629580c820d03edf561835) Add DEEPEP_SKIP_KINETO env var to bypass kineto profiling ### 📊 Changes **1 file changed** (+22 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `tests/utils.py` (+22 -0) </details> ### 📄 Description ## Summary - Add `DEEPEP_SKIP_KINETO` environment variable to bypass kineto profiling - When set to `1`, uses simple CUDA event timing instead of torch.profiler - Allows using external profilers like `nsys profile -t cuda` without conflicts ## Usage ```bash DEEPEP_SKIP_KINETO=1 nsys profile -t cuda python tests/test_internode.py ``` ## Background The test scripts use kineto (torch.profiler) for capturing kernel durations. This can conflict with `nsys profile` when users want to analyze the kernels with NVIDIA Nsight Systems. With this change, users can set `DEEPEP_SKIP_KINETO=1` to disable kineto and use nsys freely. ## Test plan - [ ] Run tests with `DEEPEP_SKIP_KINETO=1` and verify they complete - [ ] Run `nsys profile -t cuda` with this env var set Addresses #516 🤖 Generated with [Claude Code](https://claude.com/claude-code) --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-27 21:04:35 +00:00