New: Cookbooks and AI ExplanationsStep-by-Step recipes to solve problems connected to Roadmaps and Cheat Sheets. Need more details? Use AI buttons for structured and simple explanations with concrete examples throughout the whole platform.Take a look
Everyone tells me to use LoRA or QLoRA, but I don't know which one fits my GPU, or whether full fine-tuning is worth the extra cost.
What you'll have at the end
A written decision record: method chosen, GPU memory budget checked, and the fallback plan if it falls short
You need
A specific base model size already picked, or a short list of candidates, and a real number for the GPU memory you have to work with, whether that is a card you own or a rented instance's listed spec.
Not covered
Picking the exact rank, alpha, or learning rate once a method is chosen, and running the training job itself.
Leans on
Should you fine-tune, or is RAG enough?
Go there first if fine-tuning, rather than RAG, is not yet the settled choice for your task.
Pick starting values for LoRA rank, alpha, and learning rate
Once LoRA or QLoRA is the chosen method, this picks the actual rank, alpha, and learning rate to start training with.
Fix a CUDA out of memory crash with QLoRA
Go there if the real training run still runs out of memory after QLoRA looked like it should have fit.
Checked 18 Aug 2026
Part of the LLM Fine-tuning cookbook