SFT
Full Form: Supervised Fine-Tuning
Category: AI Techniques
📖 Definition
SFT trains a model on carefully labeled examples of desired behavior. Unlike RLHF, it directly shows the model correct responses rather than learning from preferences.
🔑 Key Points
- Training on curated examples of correct behavior
- Simpler than RLHF but still effective
- Often used with RLHF for best results
- Requires high-quality training data
💡 Why It Matters
SFT is a foundational technique for customizing AI. It's often the first step in creating specialized AI assistants.