Reference · Glossary

Experiment tracking

Logging **prompt, model, data, and metric** for each training or eval run so you can compare and reproduce results.

When to use

Fine-tunes, RAG tuning, and A/B tests with more than one iteration.

When not to

One-off toy scripts with no intent to compare runs.

Example

Run #42: `gpt-4o-mini`, chunk 512, recall@5 = 0.81 — tagged in MLflow or a simple JSON log.