Forked from kaiokendev/4Bit Lora Trainer Notebook.ipynb
Created
April 20, 2023 16:23
-
-
Save ausboss/af3efb317c1e110acb08772675b7d909 to your computer and use it in GitHub Desktop.
4-bit LoRA Training Notebook
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Could you possibly put together a notebook on how to generate a new model from a base model and n loras? Now that we have lora and 4bit, this opens up some really interesting possibilties; but for a 30b model, inference isn't possible on community hardware (24GB) with full context. Being able to finetune our own models and then "bake" them will be a big step forward; however as far as I can tell, there's no documentation, or at least, decent tutorials, on how to do this.