This issue, which presents a problem with how Alpaca Lora saved models, hopefully, is what is causing the problems. I also see the final adapter.bin for the finetunes is like 4000 bytes, which is not a lot, so I am fairly confident this is the problem. I will run a fine-tune with the suggested fix and should have results by tomorrow morning.
https://github.com/huggingface/peft/issues/286
This issue, which presents a problem with how Alpaca Lora saved models, hopefully, is what is causing the problems. I also see the final adapter.bin for the finetunes is like 4000 bytes, which is not a lot, so I am fairly confident this is the problem. I will run a fine-tune with the suggested fix and should have results by tomorrow morning.