Add B200 CI test workflow by csahithi · Pull Request #9604 · sgl-project/sglang

csahithi · 2025-08-25T17:34:10Z

Motivation

Add test suite to run on B200 runner for blackwell testing

Modifications

Added a new github workflow pr-test-blackwell
Updated b200 test suite (which is currently empty)

Accuracy Tests

Tested the newly added test suite on b200
per-commit-8-gpu-b200.txt

Benchmarking and Profiling

Checklist

Format your code according to the Format code with pre-commit.
Add unit tests according to the Run and add unit tests.
Update documentation according to Write documentations.
Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.

gemini-code-assist

Summary of Changes

Hello @csahithi, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request introduces initial tests for the B200 GPU runner by populating the per-commit-8-gpu-b200 test suite. The primary goal is to establish a foundational CI test workflow for Blackwell testing, ensuring that critical functionalities related to FP4 operations are validated on the target hardware.

Highlights

Expanded B200 Test Suite: The per-commit-8-gpu-b200 test suite, previously empty, has been updated to include test_fp4_gemm.py and test_fp4_quantize.py. These additions are crucial for establishing initial test coverage for the B200 runner.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in issue comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist

Code Review

This pull request aims to add a new CI test workflow for B200 runners by populating a new test suite. The changes involve adding two new test files to the per-commit-8-gpu-b200 suite in test/srt/run_suite.py. My review identified a critical issue where the paths to these new test files are incorrect, which would likely cause the new CI workflow to fail. I have provided a code suggestion to correct these paths.

kushanam · 2025-08-28T16:56:44Z

thanks! LGTM now.
cc: @zhyncs

merrymercy · 2025-08-30T00:23:29Z

the sgl-kernel tests should go to separates files (https://github.com/sgl-project/sglang/blob/main/.github/workflows/pr-test-sgl-kernel.yml). You can create a new pr-test-sgl-kernel-blackwell.yml or reuse the existing one.

Do not run things in the parent folder ../../python/sglang/test/test_fp4_moe.py. You can import or copy it under test/srt

@merrymercy Thanks for the comments. Updated the PR with these changes. Created a new pr-test-sgl-kernel-blackwell.yml - https://github.com/sgl-project/sglang/actions/runs/17411110681/job/49438550895?pr=9604

nWEIdia · 2025-09-03T20:40:34Z

For the sake of convenience: pasting the B200 job directly here: https://github.com/sgl-project/sglang/actions/runs/17411110648/job/49428467799?pr=9604

nWEIdia · 2025-09-03T20:49:59Z

https://github.com/sgl-project/sglang/actions/runs/17411110648/job/49539116290?pr=9604 this PR does not change ".github/workflows/pr-test-amd.yml"
I guess we can ignore half of the 6 failures?

nWEIdia · 2025-09-04T04:53:22Z

Hi @zhyncs , could you please help resolve the conflict and merge this? Thanks!

Signed-off-by: Sahithi Chigurupati <chigurupati.sahithi@gmail.com>

csahithi · 2025-09-04T11:36:35Z

@zhyncs I resolved the merge conflicts, could this PR pls be merged now? Thanks!

nWEIdia · 2025-09-04T16:17:31Z

          cd test/srt
          python3 run_suite.py --suite per-commit-8-gpu-deepep

-  unit-test-backend-8-gpu-b200:


@csahithi Why do we remove pr-test unit-test-backend-8-gpu-b200?

this job is actually running on B200 CI

This was separated out into pr-test-blackwell.yml

nWEIdia · 2025-09-04T16:19:43Z

@@ -0,0 +1,79 @@
+name: PR Test (Blackwell)


What is the benefit of separating the test out in a new yml file?

The idea is to add additional tests for blackwell in the future including e2e suite similar to pr-test-amd.yml

nWEIdia · 2025-09-04T16:20:30Z

+        timeout-minutes: 60
+        run: |
+          cd test/srt
+          python3 run_suite.py --suite per-commit-8-gpu-b200 --auto-partition-id 0 --auto-partition-size 1


Currently we have per-commit unit testing.
Nightly should bet testing something bigger.

Yes, similiar to above, the idea is to add a lot more tests going forward including running, for example, end to end tests. This is just the beginning to get started on that going forward.

Currently, we want some basic CI at least to be enabled on blackwell (there is no CI running on B200 nodes at the moment). This PR serves that purpose to just add basic CI testing which is why it is crucial to get this merged ASAP.

this job is actually running on B200 CI
Re: "there is no CI running on B200 nodes at the moment"

cliffwoolley · 2025-09-04T22:48:26Z

+          PIP_SUFFIX="--break-system-packages"
+          pip uninstall sgl-kernel -y $PIP_SUFFIX || true
+          pip uninstall ~gl-kernel -y $PIP_SUFFIX || true
+          pip install torch==2.8.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/test/cu129 $PIP_SUFFIX


maybe we shouldn't hardcode the torch & cuda versions here?

csahithi requested review from merrymercy and zhyncs as code owners August 25, 2025 17:34

gemini-code-assist Bot reviewed Aug 25, 2025

View reviewed changes

Comment thread test/srt/run_suite.py Outdated

csahithi force-pushed the b200-testing branch from 47a6e0e to cdefb0a Compare August 25, 2025 18:12

kushanam reviewed Aug 27, 2025

View reviewed changes

Comment thread test/srt/run_suite.py Outdated

Comment thread .github/workflows/pr-test-blackwell.yml Outdated

Comment thread .github/workflows/pr-test-blackwell.yml Outdated

Comment thread .github/workflows/pr-test-blackwell.yml Outdated

kushanam approved these changes Aug 28, 2025

View reviewed changes

merrymercy reviewed Aug 30, 2025

View reviewed changes

csahithi requested review from BBuf, FlamingoPg, HaiShaw, HandH1998, ispobock and yizhang2077 as code owners September 1, 2025 15:59

csahithi force-pushed the b200-testing branch 2 times, most recently from 8b98ca0 to 57bbcbd Compare September 2, 2025 17:23

Fridge003 self-assigned this Sep 3, 2025

zhyncs reviewed Sep 3, 2025

View reviewed changes

Comment thread scripts/ci/ci_install_dependency.sh Outdated

Add b200 testing workflow

d45aa52

Signed-off-by: Sahithi Chigurupati <chigurupati.sahithi@gmail.com>

csahithi force-pushed the b200-testing branch from 0b70562 to d45aa52 Compare September 4, 2025 11:34

nWEIdia reviewed Sep 4, 2025

View reviewed changes

cliffwoolley suggested changes Sep 4, 2025

View reviewed changes

Fridge003 mentioned this pull request Sep 8, 2025

Add disable_chunked_prefix_cache feature to TRTLLM MLA #10178

Closed

4 tasks

Fridge003 closed this Nov 3, 2025

Conversation

csahithi commented Aug 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Motivation

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

kushanam commented Aug 28, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

nWEIdia commented Sep 3, 2025

Uh oh!

nWEIdia commented Sep 3, 2025

Uh oh!

Uh oh!

nWEIdia commented Sep 4, 2025

Uh oh!

csahithi commented Sep 4, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

csahithi commented Aug 25, 2025 •

edited

Loading