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

Reported Scores in Paper may be low #13

Open
lukemcdermotttt opened this issue Mar 14, 2025 · 0 comments
Open

Reported Scores in Paper may be low #13

lukemcdermotttt opened this issue Mar 14, 2025 · 0 comments

Comments

@lukemcdermotttt
Copy link

Table 6 shows Lolcats distilled from Llama 3.1 8B gets 69.7% acc on Winogrande (w/ Llama 3.1 8B gets 73.5% acc)

Then Table 3/4 show that Lolcats distilled from Llama 3 8b gets 74.1% on Winogrande (over Llama 3 8B 73.1% acc), much better.

I reran the LM eval on huggingface checkpoints of Lolcats Llama 3.1 8b (on a tweaked repo, so mine might be wrong) and got winogrande scores that relate closer to your Llama 3 8b-version model. Maybe Table 6 scores should be re-ran to boost performance in paper?

I also might be missing something, the score just seemed odd to me when reading through the paper.

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