Pre-training and fine-tuning
A foundation model is pre-trained once, at huge cost, on a broad mix of text or images, so it can do many tasks. Fine-tuning trains it a little more on a smaller, focused set of examples, such as a company's best support replies, so it picks up a style, a format or a specialty. Fine-tuning changes the weights, so it costs more than prompting or RAG and needs good examples, but the change lasts. Reinforcement learning from human feedback (RLHF) is a kind of fine-tuning that teaches a model which answers people prefer. A rule of thumb: try prompting first, use RAG for facts that change, and fine-tune for a consistent style or task.
Why it matters: Knowing when each approach fits saves money and is a favourite exam scenario.
Builds on: Training versus using a model
Learn it, then prove it
Sign in (free) to take this step's mastery check and save your journey.
Contact Us
Questions or feedback about anything? We'd love to hear from you.
You can reach us at: support@seriousmindware.com