Fig. I — A run, in order
LoRA Deep Dive
4 parts, read in order — each one assumes the one before it. Back toall runs orthe full archive.
The parts
Full fine-tuning of a large language model is brutally expensive — not because of the forward pass, but because training multiplies memory.
- LoRA: Why Fine-Tuning a Billion-Parameter Model Only Takes a Few Million Parameters
- QLoRA: The Paper That Put Large-Model Fine-Tuning on a Desk
- The LoRA Knobs: Rank, Alpha, Targets, and the Settings Everyone Copies Without Understanding
- Life After Training: Merging LoRAs, Stacking Them, and Serving a Hundred Fine-Tunes on One GPU