Models, reasoning & training · Concept

Fine-tuning

Adapting pretrained model weights with task, domain, or preference data; supervised fine-tuning and parameter-efficient methods are distinct approaches.

Example

Choose the training objective and parameter strategy before comparing a fine-tuning run with prompting or retrieval.

The distinction that matters

Umbrella term; specify the objective and whether all or only added or selected parameters change.

Related terms

SFT · PEFT · Post-training

Primary references

Reference definition