Skip to content

convert query, key, cos, sin to bf16 if any are bf16#182

Closed
skaulintel wants to merge 2 commits into
habana-mainfrom
skaulintel/fusedrope_nanfix
Closed

convert query, key, cos, sin to bf16 if any are bf16#182
skaulintel wants to merge 2 commits into
habana-mainfrom
skaulintel/fusedrope_nanfix

Conversation

@skaulintel
Copy link
Copy Markdown

Llama training produces nan loss after fusedrope bf16 change introduced in #140. This change fixes that

@skaulintel skaulintel requested a review from a user April 30, 2024 01:59
@skaulintel skaulintel marked this pull request as draft April 30, 2024 01:59
@skaulintel skaulintel marked this pull request as ready for review May 8, 2024 16:39
@skaulintel skaulintel closed this Jun 13, 2024
@skaulintel skaulintel reopened this Jun 13, 2024
@skaulintel
Copy link
Copy Markdown
Author

huggingface#1072

@skaulintel skaulintel closed this Jul 9, 2024
astachowiczhabana pushed a commit that referenced this pull request Mar 6, 2025
Author:    Jay Gala <jaygala@habana.ai>

Co-authored-by: root <root@jaygala-vm-u22.habana-labs.com>
astachowiczhabana pushed a commit that referenced this pull request Mar 17, 2025
Author:    Jay Gala <jaygala@habana.ai>
astachowiczhabana pushed a commit that referenced this pull request Mar 31, 2025
Author:    Jay Gala <jaygala@habana.ai>

Co-authored-by: root <root@jaygala-vm-u22.habana-labs.com>
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.

1 participant