Training types
Training types determine how model weights are updated during customization. Every customization technique (SFT, DPO, RFT) can be combined with either training type. Choose based on your quality requirements, compute budget, and use case.
- LoRA — Low-Rank Adaptation
Trains a small set of adapter weights. Lower cost, faster training. Required for continuous customization.
- FFT — Full Fine-Tuning
Updates all model weights. Higher compute, deeper customization. Best for large domain shifts.