LoRA Learns Less and Forgets Less Updated

Updated research paper showing alpha = 2*rank for LoRA finetuning of LLMs gets much better results + other cool results!