Atlas · GenAI 2026
Supervised Fine-Tuning (SFT)
Training LLMs on (instruction, response) pairs with cross-entropy loss
conceptPeak: 2023Fine-Tuning MethodsAI consensus: 3/3
Prerequisites
No prerequisites.
Recommended reference
Zhou et al. (2023) 'LIMA: Less Is More for Alignment' — 1000 examples > millions
Notes from AI deep research
Anthropic Opus
Foundational fine-tuning step. LIMA paper: jakosc danych > ilosc. Wymaga dyscypliny w formatowaniu
Related skills
- → is subcategory of: LLM Fine-Tuning(3/3)