genuine question : Why ?
#1
by
mammour
- opened
MistralAI :
"We got you pals, there, take this, it's a beautiful pair of system prompt tokens"
fine-tuners :
Based on training of test models I found the difference in training loss and trial usage to be negligible at best. Both formats are fine, but since the PersonalityEngine series is standardized on ChatML I felt it best to prioritize making it a drop in replacement.
Thank you for your feedback.
PocketDoc
changed discussion status to
closed