The Skill Explained
What Is Fine-Tuning?
Fine-tuning is training an already trained model a little more on a smaller, focused dataset so it gets better at one task, style or format. It changes the model's weights. It suits teaching a behavior, such as a tone or a strict output format, more than adding new facts.
Why AI Employers Ask for It
Prompts and retrieval solve many problems, but some tasks need a model that has learned a pattern deeply: a strict output format, a specialised vocabulary, a smaller and cheaper model that matches a big one on one task. Employers ask for Fine-Tuning in machine learning and research roles, and in teams that run their own models.
How It Works
You start from a pre-trained model and a dataset of input and output pairs that show the behaviour you want, often hundreds to thousands of high-quality examples. Training adjusts the model's weights so its outputs move closer to your examples.
Full fine-tuning updates all weights and needs a lot of memory. Parameter-efficient methods such as LoRA train only small added layers, which makes it possible to adapt large models on modest hardware and keep several adaptations of one base model. Instruction tuning and preference tuning (for example RLHF or DPO) are post-training stages that teach a model to follow instructions and prefer better answers.
Data quality decides the result. Teams hold back test examples, compare the tuned model with the base model and with a good prompt, and watch for forgetting general skills or memorising sensitive data.
What You Do With It at Work
- Decide whether a problem needs fine-tuning or whether prompting and RAG are enough.
- Build and clean a training dataset of examples.
- Run parameter-efficient fine-tuning (for example LoRA) and track experiments.
- Compare the tuned model with the base model on a held-out test set.
- Deploy and monitor the tuned model, and plan retraining.
Tools and Terms You Will Meet
- Pre-trained model
- A model already trained on large general data, used as a starting point.
- LoRA
- A method that fine-tunes a few small added weights instead of the whole model.
- Instruction tuning
- Training on instruction and answer pairs so a model follows requests.
- Catastrophic forgetting
- When new training makes a model lose skills it had.
- Held-out set
- Examples kept out of training to test the model fairly.
Industry Example: Pharma
A company fine-tunes a small open model to extract dosing, route and frequency from prescriptions into a fixed format, so it runs on internal servers at low cost instead of sending data to an external service.
A Common Mix-Up
Fine-tuning is usually not the best way to add knowledge, such as new documents. For knowledge, retrieval (RAG) is easier to update; fine-tuning is better for behaviour, format and style.
How Deep Do You Need to Go?
- Know itExplain when fine-tuning helps and when prompting or RAG is better.
- Use itFine-tune a small model with LoRA and evaluate it against a baseline.
- Master itDesign post-training data and pipelines, including preference tuning, at scale.