Fine Tune Training
#2
by
DazzlingXeno
- opened
How did you fine tune this? Did you convert the Gutenberg dataset to mistral instruct format or did you just use a JSON/parquet? Thanks in advance.
DazzlingXeno
changed discussion title from
Training
to Fine Tune Training
I used a modified version of Maxime Labonne's ORPO notebook. The data was formatted using ChatML. The changes are shown in this thread: https://huggingface.co/nbeerbower/mistral-nemo-gutenberg-12B-v2/discussions/1
Thank you 😊
I'm thinking of using the new command-r 32 or magnum 34b. Leaning towards Magnum as that's already ChatML. So I'm going to have to test your model on both formats to see if it matters.