Files
20d70d9958 Add dataset cache / mixing support (#522)
* Add dataset cache / mixing support

* prototyping

* makesure to shuffle with seed

* quick fix

* add finetune1

* support preference tuning as well

* refactord

* allow passing in an SFT message key

* allow customizing chosen / rejected key

* fix tests

* add huggingface card

* refactor

* quick fix

* add an option to cache dataset only

* Add some logic for caching.

* make mason work with the latest change

* Use the latest dataset caching logic

* restore change

* push docs

* update docs

* quick update

* quick push

* push

* Just replace the existing dpo / finetune

* remove unused files

* Apply suggestions from code review

Co-authored-by: Nathan Lambert <[email protected]>

* quick change

* update docs and fix mason

* quick update on docs

* use the default entity

* format

---------

Co-authored-by: Nathan Lambert <[email protected]>
2025-01-28 09:02:38 -05:00
..