Skip to content

Conversation

@mgoin
Copy link
Member

@mgoin mgoin commented Sep 23, 2025

Purpose

Due to compiling full cudagraphs by default now in #25444, gpt-oss crashes on Hopper.

vllm serve openai/gpt-oss-20b
...
(EngineCore_DP0 pid=2061997)   File "/home/mgoin/code/vllm/vllm/v1/worker/gpu_model_runner.py", line 3503, in create_attn_groups
(EngineCore_DP0 pid=2061997)     attn_metadata_builders.append(attn_backend.get_builder_cls()(
(EngineCore_DP0 pid=2061997)                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=2061997)   File "/home/mgoin/code/vllm/vllm/v1/attention/backends/flash_attn.py", line 204, in __init__
(EngineCore_DP0 pid=2061997)     raise ValueError(
(EngineCore_DP0 pid=2061997) ValueError: Capture size larger than 992 is not supported for full cuda graph.

Test Plan

Test Result

vllm serve openai/gpt-oss-20b crashes on main but succeeds with this PR


Essential Elements of an Effective PR Description Checklist
  • The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
  • The test plan, such as providing test command.
  • The test results, such as pasting the results comparison before and after, or e2e results
  • (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
  • (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

@mgoin mgoin added the bug Something isn't working label Sep 23, 2025
@mergify mergify bot added the gpt-oss Related to GPT-OSS models label Sep 23, 2025
Copy link
Contributor

@gemini-code-assist gemini-code-assist bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request correctly addresses a crash on Hopper GPUs for the gpt-oss model by lowering the maximum CUDA graph capture size to 992, which is the compatibility limit for FlashAttention v3. The change is well-targeted and effective. I have one suggestion to improve the code's maintainability by replacing the hardcoded value 992 with a named constant.

@vllm-bot vllm-bot merged commit a8ffc4f into vllm-project:main Sep 23, 2025
9 of 14 checks passed
@github-project-automation github-project-automation bot moved this from To Triage to Done in gpt-oss Issues & Enhancements Sep 23, 2025
FeiDaLI pushed a commit to FeiDaLI/vllm that referenced this pull request Sep 25, 2025
yewentao256 pushed a commit that referenced this pull request Oct 3, 2025
gjc0824 pushed a commit to gjc0824/vllm that referenced this pull request Oct 10, 2025
xuebwang-amd pushed a commit to xuebwang-amd/vllm that referenced this pull request Oct 10, 2025
choprahetarth pushed a commit to Tandemn-Labs/vllm that referenced this pull request Oct 11, 2025
lywa1998 pushed a commit to lywa1998/vllm that referenced this pull request Oct 20, 2025
xuebwang-amd pushed a commit to xuebwang-amd/vllm that referenced this pull request Oct 24, 2025
rtourgeman pushed a commit to rtourgeman/vllm that referenced this pull request Nov 10, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bug Something isn't working gpt-oss Related to GPT-OSS models

Projects

Status: Done

Development

Successfully merging this pull request may close these issues.

2 participants