Fine-tuning, LoRA and instruction tuning
From a base model to an assistant: supervised instruction tuning, loss masking, why full fine-tuning is expensive, LoRA's low-rank update with a tested merge identity and parameter arithmetic, QLoRA, hyperparameters and the common failure modes.
17 min·Verified 2026-111 animated diagramPro+50 XP
A base model is trained to continue text. An assistant is trained to follow instructions. The step between them, and every later customisation, is fine-tuning.
Pro content
Keep reading with Inference Academy Pro
Unlock the rest of this lesson, including 1 animated diagram, plus every advanced lesson, hands-on lab, the larger playground models and completion certificates.