Modern peft methods with LoRA actually do reduce training time by orders of magnitude. Here's an example of 20 seconds per epoch on a single consumer GPU: http

Modern peft methods with LoRA actually do reduce training time by orders of magnitude. Here's an example of 20 seconds per epoch on a single consumer GPU: https://github.com/johnsmith0031/alpaca_lora_4bit/issues/7#issue-1635228539 https://github.com/johnsmith0031/alpaca_lora_4bit/issues/7#i...