OpenAssistant/pythia-12b-sft-v8-7k-steps is now also available

[andreaskoepf]

Super cool demo ... !

Just wanted to mention that our SFT-8 pythia-12b is now also available: OpenAssistant/pythia-12b-sft-v8-7k-steps. Outputs of both SFT-4 & SFT-8 side-by-side can be viewed here.

The data-mixture that was used for both fine-tuning runs differs a bit (SFT-8 was trained on more data and more recent instruction datasets). In the end it is a matter of taste which one you like better. Some people complain that sft-8 doesn't want to help building bombs or harm humans - whether that is good or bad is of course debatable. I recommend to play with both SFT-4 & 8 and then decide for yourself ... :-)