10000 vocab : fix ugm tokenizer precision by CISC · Pull Request #13743 · ggml-org/llama.cpp · GitHub
[go: up one dir, main page]

Skip to content

vocab : fix ugm tokenizer precision #13743

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

Merged
merged 1 commit into from
May 24, 2025
Merged

Conversation

CISC
Copy link
Collaborator
@CISC CISC commented May 24, 2025

Under certain conditions storing the score_sum as float leads to imprecision and incorrect token selection.

Fixes #13725

@CISC CISC requested a review from ggerganov May 24, 2025 09:44
@CISC CISC mentioned this pull request May 24, 2025
6 tasks
@CISC CISC merged commit c3a2624 into master May 24, 2025
46 checks passed
@CISC CISC deleted the cisc/fix-ugm-tokenizer-precision branch May 24, 2025 10:29
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

Successfully merging this pull request may close these issues.

Eval bug: UGM tokenizer sometimes outputs wrong tokens/in the wrong order
2 participants
0