cog-llama-template
cog-llama-template copied to clipboard
Why are there two training functions?
I see two functions called train defined in different places. Which one is the user-facing training interface?
- train.py is the thing cog.yaml is pointing to.
- training/trainer.py is ... maybe being used by the former?
To add to the confusion, the llama training inputs on the site don't seem to match the inputs of either of the train functions:
To Do
- [ ] If they're both used, clarify the purpose of each with a docstring or comment
- [ ] If they're not both used, get rid of the dead one
cc @replicate/models
Should be looking at the train-dev branch (but everything might still be broken in that branch too)