Workaround for RoPE computed in bf16 for GPT-NeoX#746
Merged
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
libinta
approved these changes
Mar 12, 2024
gplutop7
pushed a commit
to HabanaAI/optimum-habana-fork
that referenced
this pull request
Oct 15, 2025
…uggingface#2295) (huggingface#746) align evaluate version question answering 0.4.5 Co-authored-by: Grzegorz Pluto-Prondzinski <gplutopx@habana.ai>
gplutop7
pushed a commit
to HabanaAI/optimum-habana-fork
that referenced
this pull request
Nov 6, 2025
…uggingface#2295) (huggingface#746) align evaluate version question answering 0.4.5 Co-authored-by: Grzegorz Pluto-Prondzinski <gplutopx@habana.ai>
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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?
When using DeepSpeed, the whole model is casted to bf16, including RoPE parameters. This makes the trained model performance much worse so this PR ensures RoPE parameters are computed in fp32.
Before submitting