Tool
Which fine-tuning method should you use?
Twenty-nine methods — adapters, memory tricks, alignment and multi-GPU strategies — with what each one trades away and where it actually lives. Answer two or three questions if you'd rather be pointed straight at one.
Decision wizard
Narrow it down by constraint.
Covers the common paths, single- and multi-GPU. The niche methods — rsLoRA, LoftQ, OFT, AdaLoRA, IA³, MoD, soft prompting — are in the full list below.
All methods
Or browse the lot.
Running these for real
Most of this was learned on 32 RTX 3090s.
The VRAM notes come from actually hitting those limits. If you're picking a method for a real project, I'm happy to talk it through.