SetFN
SetFNDocs2026년 9월 18일

Fine-tuning Introduction

The Fine-tuning plugin, built on LLaMA Factory, trains a local LLM (including small language models such as sLLM) to fit the domain you want. You can handle the entire process — from preprocessing raw data, through training, evaluation, and chat testing, to exporting the resulting model — all from a single screen.

Preprocessing training data

Workflow

Fine-tuning proceeds through the following five stages.

StageDescription
PreprocessingConverts and cleans raw data (XML, MD, HTML, JSON, TXT) to fit the training purpose (PT/SFT)
TrainingRuns fine-tuning by specifying a dataset, model, and training method (STAGE)
EvaluationReviews the results of a completed training run and re-validates dataset validity
Chat TestLets you chat directly with the trained model to check the results
ExportManages training runs and connects successful weights as the Base Model for the next fine-tuning round

Good to Know

  • Training requires GPU resources. Check your server specifications in advance.
  • Once started, training continues in the background until you stop it.