Skip to content

[WIP] FIX Make Mixtral LoRA loading work#44478

Open
BenjaminBossan wants to merge 6 commits intohuggingface:mainfrom
BenjaminBossan:peft-weight-conversion-fixes
Open

[WIP] FIX Make Mixtral LoRA loading work#44478
BenjaminBossan wants to merge 6 commits intohuggingface:mainfrom
BenjaminBossan:peft-weight-conversion-fixes

Conversation

@BenjaminBossan
Copy link
Member

@BenjaminBossan BenjaminBossan commented Mar 5, 2026

Required fixes:

  • some code was using unordered data structures, making weight order random
  • adjust alpha to offset increased rank from fusion
  • import functions from PEFT if available

See huggingface/peft#3083.

Required fixes:

- some code was using unordered data structures, making weight order
random
- adjust alpha to offset increased rank from fusion
- import functions from PEFT if available
BenjaminBossan and others added 2 commits March 6, 2026 16:00
This can be used in PEFT to apply weight conversion there without having
to either re-implement the whole weight conversion machinery or having
to call transformer_model.load_adapter. We want to avoid the latter
because there is a lot of PEFT custom logic to weight loading which
would require adjusting for that case.
@BenjaminBossan BenjaminBossan marked this pull request as ready for review March 6, 2026 16:17
@BenjaminBossan
Copy link
Member Author

Note: The added test is a @slow test, probably need to invoke some magic command to run those.

mark stuff that can be removed with peft 0.19
…san/transformers into peft-weight-conversion-fixes
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