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.

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

Gemma2 BOS token for likelihood_rolling #2382

Open
MFajcik opened this issue Oct 4, 2024 · 0 comments
Open

Gemma2 BOS token for likelihood_rolling #2382

MFajcik opened this issue Oct 4, 2024 · 0 comments

Comments

@MFajcik
Copy link

MFajcik commented Oct 4, 2024

Hi,

creating an issue to track progress on optional BOS prefix in all windows for llh_rolling methods. This is useful for models such as Gemma2 family.

Based on original issue reported in discord

I investigated some strange numbers on language modeling tasks (evaluated with ppl), and I assume BOS is added just once (not in subsequent frames of loglikelihood_rolling, when document is windowized). Is this the case? And shouldn't BOS be added for gemma everytime?

I stopped model call randomly in debugger and verified its inputs to see, that BOS is actually not there in llh_rolling everytime

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant