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

vLLM bug #83

Closed
hqhQAQ opened this issue Feb 13, 2025 · 2 comments
Closed

vLLM bug #83

hqhQAQ opened this issue Feb 13, 2025 · 2 comments

Comments

@hqhQAQ
Copy link

hqhQAQ commented Feb 13, 2025 •
edited
Loading

Current version of vLLM training may have bug.
For example, when GPU number and num_generations are both 4, current vLLM training requires that in each batch, all 4 samples correspond to the same question and image. However, in practice, although RepeatRandomSampler is used, these 4 samples may correspond to different questions and images.
(Correct me if I am wrong)
@TobiasLee
Copy link
Collaborator

can you provide outputs for the different questions and images?
if so, the original GRPOTrainer would be wrong as well: huggingface/trl#2776

@hqhQAQ
Copy link
Author

hqhQAQ commented Feb 13, 2025

可能我搞错了, 我再看看😰

@hqhQAQ hqhQAQ closed this as completed Feb 13, 2025
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

No branches or pull requests
2 participants
@TobiasLee @hqhQAQ