Fix Falcon weight mapping for H2O.ai checkpoints#953
Merged
Narsil merged 1 commit intohuggingface:mainfrom Aug 31, 2023
Merged
Conversation
During the safetensor conversion, duplicate weights are removed. However, which of the duplicates gets removed, differs per checkpoint. In some, like `h2oai/h2ogpt-oig-oasst1-falcon-40b`, the weight `transformer.word_embeddings.weightSafetensor` gets removed. In others, `lm_head.weight` gets removed. Long story long, we need to support both. Originally, f018143 mapped `lm_head` to `word_embeddings`. Then ac736fd switched this around. This commit merges them and allows for both.
Contributor
|
@Vinno97 The supposedly randomness should have been removed by earlier commits (using ttransformers hints). |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What does this PR do?
During the safetensor conversion, duplicate weights are removed. However, which of the duplicates gets removed, differs per checkpoint. In some, like
h2oai/h2ogpt-oig-oasst1-falcon-40b, the weighttransformer.word_embeddings.weightSafetensorgets removed. In others,lm_head.weightgets removed. Long story long, we need to support both.Originally, f018143 mapped
lm_headtoword_embeddings. Then ac736fd switched this around. This commit merges them and allows for both.Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
@Narsil, you wrote both commits I referenced in this PR. I think you'll understand this change :)