Decoder. plain-English AI glossary

fine-tuning API

● Corefine-tuning API

An API endpoint where you upload training data and request the model to fine-tune on it, returning a new model variant optimized for your task.

Think of it like

Sending your generic car to a shop and saying 'make it handle better on mountain roads'—you get a custom variant back.

Example

Upload 500 examples of customer emails + correct support-ticket categories. Call the fine-tuning API. In an hour, get back a model that's better at categorization than the base model.

How it actually works

Fine-tuning APIs run the training loop on their hardware, charge by token or example, and give you a model checkpoint to call. Tradeoff: you don't see training loss curves, you can't customize loss functions, but you're not managing GPUs. Popular for task-specific optimization when in-context examples don't cut it.

For product teams

Improves model quality on domain tasks; more cost-effective than prompt engineering beyond a certain threshold.

For engineers

Managed training; no GPU setup; token-based pricing.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open DecoderAdd to Chrome