Adapt a model
Adjust a pretrained model for a new task or domain using additional training.
Desired outcome
A task-adapted model or adapter with documented training conditions.
Explore from the outcome
Where this function can lead
These documented connections are starting points for selection, not a ranked list.
Ways to do it
Methods that can help achieve this outcome.
- Low-rank adaptation (LoRA)
Train low-rank adapter matrices for selected model weights.
Link evidence: PEFT methods overview
- Parameter-efficient fine-tuning
Train a limited set of parameters or adapters while keeping most base weights fixed.
Link evidence: PEFT methods overview
Tools and services
Software or services that can put a method to work.
- Hugging Face PEFT
Python library for parameter-efficient model adaptation, including LoRA.
Link evidence: Hugging Face PEFT repository
Typical inputs
- Base model, training examples, and adaptation configuration
Typical outputs
- Updated weights or adapter artifact
Sources
Last reviewed 2026-09-30