run_lora_clm.py support for other datasets

[tmabraham]

Feature request

right now the script is hardcoded for either "tatsu-lab/alpaca" or "timdettmers/openassistant-guanaco" and using any other dataset throws an error.

Motivation

It would be nice to be able to finetune on our own datasets.

Your contribution

Happy to test any code out for this...

[regisss]

Yes, I agree that would be much better.
Can you share a command line that fails? And the error message you get please?

[dmsuehir]

I posted PR #955 that fixes this issue and will allow other datasets to be used with run_lora_clm.py. I tried it with several different datasets to check the functionality, but please let me know if there's a specific one that you'd like me to try.