LoRA explained (and a bit about precision and quantization)