Hey all, I spent some time over the last few days putting together a library to make fine tuning easier under specific situations. You're using a GPU, or multiple GPUs on a single box. This comes set up with accelerate for multi gpu training out of the box and support PEFT and Lora, 8bit and a few other things.You just need to import Pipeline from `ditty.pipeline` and overwrite the dataset method to return a dataloader setup with your dataset of choice. And then instantiate your pipeline and call `run`