Format? Base finetune?
#1
by
brucethemoose
- opened
Is this just your own finetune?
Is the prompting syntax Puffin? Or llama 2 chat?
I borrowed the sample datasets from the Thespin models for this initial test fine-tune. You can try using the format detailed here:
https://huggingface.co/cgato/Thespis-13b-v0.4
I did not use any format and just left the default for instruct in ooba. I set the max context down to 32K for testing and only changed this setting from ooba defaults:
