🌊Introducing DPO Mix 7K, a small DPO dataset that does wonders!
Yesterday,
@_philschmid
&
@_lewtun
showcased its strength with Zephyr Gemma
If you're looking for a small, diverse, high quality DPO dataset check it out!
We built it filtering & mixing our recent DPO datasets:
- Reranked
@intel
Orca pairs
- Cleaned
@openbmb
UltraFeedback
-
@ldjconfirmed
Capybara DPO
It worked very well for us (CapybaraHermes) & others, but never introduced it publicly
Now it's time: It deserved its own image!