[VLM] Fixing request timeout error, and enable VllmDeployer to fail fast if the underying vllm serve process already failed
#2409
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Contributor
|
MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅ |
johncalesp
approved these changes
Dec 10, 2025
Contributor
johncalesp
left a comment
johncalesp
left a comment
There was a problem hiding this comment.
Looks good, thank you!
johncalesp
reviewed
Dec 10, 2025
Victor49152
approved these changes
Dec 11, 2025
Victor49152
left a comment
Victor49152
left a comment
There was a problem hiding this comment.
Verified for a few runs and works good
…rf-inference into wangshangsam/fix-req-timeout
… need to download the model for the fisrt time.
Contributor
Author
|
This PR is ready. @mrmhodak @hanyunfan Could you help to approve and merge this PR? |
…rf-inference into wangshangsam/fix-req-timeout
…condhand f1 score
nvzhihanj
reviewed
Dec 16, 2025
nvzhihanj
reviewed
Dec 16, 2025
nvzhihanj
approved these changes
Dec 16, 2025
nvzhihanj
reviewed
Dec 16, 2025
…rf-inference into wangshangsam/fix-req-timeout
…rf-inference into wangshangsam/fix-req-timeout
nvzhihanj
reviewed
Dec 17, 2025
…rf-inference into wangshangsam/fix-req-timeout
…rf-inference into wangshangsam/fix-req-timeout
Contributor
Author
|
@nvzhihanj do you wanna take another look? If it LGTY, we can merge it. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR fixes two issues:
DefaultAioHttpClientinAsyncOpenAI, it turns out thatDefaultAioHttpClient's HTTP request timeout setting would override theAsyncOpenAI's chat completion request timeout setting, thereby, I also need to set the timeout setting inDefaultAioHttpClientas well.VllmDeployer, it would still wait 20 mins (or whatever the timeout is) to fail. Now it can fail fast as soon as the underyingvllm serveprocess fail.In addition, this PR also:
3. Introduce example slurm scripts that demonstrates how to run the benchmark in a GPU cluster managed by slurm.
4. Change the default
server_expected_qpsandserver_target_latencyto the value that we are proposing: