allenai / allenai/olmes

newer transformers library versions break olmo3, fix provided in a PR

Open
#54 2 comments 1 reaction 0 assignees View on GitHub
Dominant language
Python
Stars
395
Forks
105
PR merge metrics
No merged PRs in 30d

Description

I submitted the following PR against transformers: https://github.com/huggingface/transformers/pull/45945

OLMo 3 scaled RoPE handling was broken by a some sweeping change refactoring RoPE handling. As a result, using a model such as `allenai/Olmo-3-1025-7B` produces incorrect outputs. I tested this by measuring the Olmo3 paper table 3 results and can see that using f.ex. transformers 5.5.3, the table 3 cannot be reproduced. However, I re-measured with my fix and the results seem to be fixed if I use a transformers version that includes my fix.

Would anyone from Ai2 be willing to review the change in the above PR?

Contributor guide

No contributing guide indexed for this repository

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.