visp lab

Guided LoRA Demo

Non-technical walkthrough: inspect train/valid data, run a real tiny-model LoRA, then test the model immediately and compare what changed.

Open training slides Open Game Builder Compare

Glossary: Train data teaches patterns. Validation data checks whether the model learned something useful on examples it did not see during training.

Step 1: Understand and customize data

Keep examples short and consistent. Tiny models train quickly, but quality improves when prompts and completions follow one clear format.

Sample data packDemo model

Task request: rewrite this whole dataset to a new behavior

Uses gpt-4.1-nano. Requires OPENAI_API_KEY.

Train data (teaches model)

Row 1

Row 2

Row 3

Validation data (checks generalization)

Row 1

Row 2

Stage data to unlock Step 2.

Active adapter: none activated yet