Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Non-determinism in evaluate.predict when using vllm #145

Open
jacob-morrison opened this issue Apr 10, 2024 · 0 comments · May be fixed by #151
Open

Non-determinism in evaluate.predict when using vllm #145

jacob-morrison opened this issue Apr 10, 2024 · 0 comments · May be fixed by #151

Comments

@jacob-morrison
Copy link
Contributor

Conversation in slack: https://allenai.slack.com/archives/C06GS4HAWJV/p1712773278326539

Similar issue: vllm-project/vllm#966

We saw non-deterministic model responses when doing greedy decoding using evaluate/predict.py, and disabling vllm fixed this issue. Two potential fixes: upcast models to float32, and upgrade vllm to a later version
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

Update all versions for new OLMo models (and others) allenai/open-instruct
1 participant
@jacob-morrison