No result difference after QLoRa

artidoro / qlora

QLoRA: Efficient Finetuning of Quantized LLMs

https://arxiv.org/abs/2305.14314

MIT License

10.08k stars 823 forks source link

No result difference after QLoRa #155

Open anshumantanwar opened 1 year ago

anshumantanwar commented 1 year ago

I tried to use qlora on dolly 2.0 3B for specs identification using dataset of 2000 items. Although qlora finetuning and inferencing is not giving any error but results are exactly same as original dolly 3b model. Is it common behavior?

artidoro commented 1 year ago

If you are getting exactly the same results it might be that you are not loading the correct parameters when doing inference and that you are actually just using the same base model without adapters.

zlh1992 commented 1 year ago

do you solve the problem? I just meet the same problem. the loss of eval_data_set is very low, but I got the same result, when I use the alpaca generate.py script to infer the lora result.