Difference with v1
#1
by
EloyOn
- opened
How did this model turn out with the extra LoRA training? Better than the first one?
I can't try them myself, 32b is out of my reach at the moment.
I haven't done extensive comparison between the 2 yet. Honestly I wouldn't be surprised if the difference is minimal, but the higher rank LORA and learning rate improve the reward/training metrics a lot.
Thanks for the reply, I'll download this version instead then (for when I'm able to run it).
EloyOn
changed discussion status to
closed