fix: bump SGLang version to 0.5.14#128
Conversation
Signed-off-by: Doğaç Eldenk <dogacel@gmail.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d527c2d107
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| SGLANG_VERSION="${SGLANG_VERSION:-v0.5.10.post1}" | ||
| SGLANG_COMMIT=94f03a39dbd39edfc2b118b5357bbbadaaa9ad28 | ||
| SGLANG_VERSION="${SGLANG_VERSION:-v0.5.14}" | ||
| SGLANG_COMMIT=4289f36ef960fad8268a6b94935686e792a81432 |
There was a problem hiding this comment.
With the default sglang backend, a fresh setup clones https://github.com/sgl-project/sglang.git and then checks out this hard-coded SHA. The public v0.5.14 release/tag is 49e384c, and this 4289f36... commit is not reachable from that repo, so git checkout $SGLANG_COMMIT fails before the new patch is applied. Please pin this to the actual upstream commit/tag used to generate the patch, or fetch the ref that contains it.
Useful? React with 👍 / 👎.
| @@ -0,0 +1,28 @@ | |||
| ARG SGLANG_IMAGE=lmsysorg/sglang:v0.5.14-cu130 | |||
There was a problem hiding this comment.
After the default moved to v0.5.14, tools/apply_sglang_patch.sh and tools/update_sglang_patch.sh both read this Dockerfile and grep for ARG SGLANG_COMMIT before doing anything else, but this new Dockerfile only defines SGLANG_IMAGE. Running either helper with defaults exits with Could not find SGLANG_COMMIT, so maintainers can no longer apply or regenerate the default SGLang patch unless they override the version. Add the commit arg here or update the scripts to read the commit from the new source of truth.
Useful? React with 👍 / 👎.
2 participants
Overview
Bumps SGLang version to 0.5.14 along with minor setup fixes
Motivation
Training drafters for Qwen3.5 with SGLang requires the latest version. Currently vLLM doesn't support DFlash for Qwen3.5 family, so the only engine with stable DFlash for Qwen3.5 is SGlang.
Testing
Recreate the environment + launch "examples/qwen3-8b-single-node/run.sh"
Will follow-up with DFlash recipe for Qwen3.5/6 soon.