-
Notifications
You must be signed in to change notification settings - Fork 508
Complete type checking for OLMo support (builds on #816) #1081
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
base: dev-3.x
Are you sure you want to change the base?
Conversation
Originally from TransformerLensOrg#718.
Fix to OLMo 2 normalization
|
@jlarson4 do you know when this might be merged? |
Hi @taziksh! Sorry for taking so long to get back to you. The plan at present is to see if we can get the OLMo models incorporated into the 3.x beta. Bumping the minimum python version and updating jaxtyping are coming as part of 3.x, so we wanted to make sure this PR, which also includes those changes, is integrated in the same place. I am updating the base branch now and will let you know when it is incorporated into 3.x Thanks! |
# Conflicts: # debugging/hf-tl-logit-comparator.ipynb # demos/ARENA_Content.ipynb # demos/Colab_Compatibility.ipynb # demos/Main_Demo.ipynb # demos/Othello_GPT.ipynb # poetry.lock # pyproject.toml # transformer_lens/HookedEncoder.py # transformer_lens/HookedTransformer.py # transformer_lens/components/abstract_attention.py # transformer_lens/config/HookedTransformerConfig.py # transformer_lens/loading_from_pretrained.py # transformer_lens/utils.py
Description
This PR builds on #816 and resolves all remaining type checking issues, allowing the OLMo implementation to pass mypy validation.
Changes:
create_alibi_multipliers(head_idx → n_heads)Type of change
Checklist: