Support for QLORA
Hello, Can we finetune model using qlora with LMFlow? (Cf: https://github.com/artidoro/qlora) Thanks
Thanks for your interest in LMFlow! Currently we haven't supported qlora yet, but it is in our schedule. By the way, we encourage any contributions to accelerate the develop progress. Thanks very much 😄
Sorry, i've been offline the last week. I'll try to have a look for QLORA integration, but i need to check more LMFlow codebase first (i'm new to it). If anyone want to help in this effort, I think we can take example on that PR, should be something pretty similar: https://github.com/tloen/alpaca-lora/pull/487/files
Hi, thanks. Agree. I feel that we can do very similar modifications to our code. If you have any questions about our repo and need help, feel free to let us know.
interested in this work, 资瓷。
I have created a PR for support for QLoRA (Added QLoRA support for Decoder transformers with tune_strategy "Normal" #613)
Once it is reviewed hopefully it get's merged.
It has been merged a few weeks ago. Thanks for @ankitpasi 's contribution!