search

replicate / lora-training

LoRA training model packaged with Cog
MIT License
112 stars 29 forks source link

Fix/params #5

Closed cloneofsimo closed 1 year ago

cloneofsimo commented 1 year ago

Some modification on arguments that are redundant in usage / too trivial / unused. Made descriptions up-to-date with https://replicate.com/replicate/lora-advanced-training.

anotherjesse commented 1 year ago

@cloneofsimo @andreasjansson this look good to me.

I wonder if saying "every increase of rank doubles the file size?" (or that could be in the readme...)

The order of the fields in the def predict sets the order in the demo form. Would it be useful to bring template / face stuff / rank to the top under instance_data?