[ Fine-tuning ]
Fine-tuning
When a model trained for your task beats a general one, and what you get from a training project with Kontinent.
Kontinent trains models for one specific task of yours: extracting fields from a document, classifying it, or making a decision whose correct answer can be written down and checked. A small model trained on that task is often more accurate, faster and cheaper than a large general model.
Fine-tuning runs as a project with you. There is no training API and no training job in the console yet: the Training and Datasmith pages in the console are previews. To start, book a call.
When it makes sense
Fine-tuning is worth it when all of these hold:
- The task repeats often, with the same kind of input each time.
- You can write down the correct answer for a sample of documents.
- A general model through the gateway already does the task, but not well enough, not fast enough, or not cheaply enough.
If you are still working out what the task is, start with prompting a model through the gateway. The requests you send there show what the task looks like.
What you get
- The trained model as an adapter in the standard safetensors format. It belongs to you and runs with common open tooling, without Kontinent.
- Measured results: how well the base model did your task before training and how well your model does it after, on the same documents.
- A model card and an evaluation log with every test document, the answer before and after, and your correct answer.
Running the trained model
You can run the adapter yourself, or Kontinent runs it for you on dedicated GPUs behind an OpenAI-compatible endpoint.