v0.10.0 of Peft breaks finetune.py

[umm-maybe]

See title.

Temporarily, adding "peft==0.9.0" to requirements.txt and ignoring the readme.md instructions to install from git works around this issue. However, it would be better if huggingface/bigcode could coordinate between the two projects so that LoRA fine-tuning this version of starcoder remains possible.