kayetanaModels that speak your language.
Some tasks come up every single day. Fine-tuning teaches a model how your team handles them, using your own conversation examples or labeled texts, right on your server.
What you can train.
Your own chat model
Fine-tune qwen3:4b-instruct with LoRA on your conversation examples. The result is your own model, “ft-name.”
Your own classifier
Turn a labeled dataset into a classifier, “clf-name,” built on an embedding model.
Stays in-house
Training data and finished models live on your server, not with a cloud service.
How your own chat model comes together.
Collect examples
Good conversations from everyday work, like your support team's replies to common questions.
Upload a dataset
Upload the examples as a chat fine-tuning dataset.
Start training
kayetana fine-tunes qwen3:4b-instruct with LoRA.
Use the model
Your new model, “ft-name,” is ready in the Playground and through the API.
Your categories, trained in.
Once your categories are set, like the topics in your ticketing system, a dedicated classifier makes sense. Train it once from a labeled dataset, then use it in the Playground and through the API like any other classifier.
- One or more labels per text
- Built on an embedding model
- Confidence per label in the result
Chat model or classifier?
| Chat fine-tuning | Classifier | |
|---|---|---|
| Result | Your own model, “ft-name” | Classifier, “clf-name” |
| Built on | qwen3:4b-instruct | Embedding model |
| Training data | Conversation examples | Labeled texts |
| Typical use | Tone, structure and content of replies | Sorting tickets and emails |
Turn your examples into your own model.
In a demo, we'll show you how fine-tuning works with your data.
Frequently asked questions
Which model can I fine-tune?
Currently qwen3:4b-instruct, with LoRA. Classifiers are built on an embedding model.
What data do I need?
For a chat model, conversation examples as a chat fine-tuning dataset. For a classifier, texts with one or more labels.
Where does my training data live?
On your server. On the privacy page, you set a retention period for datasets, after which they're deleted automatically.
How do I use the finished model?
It shows up as “ft-name” or “clf-name” and is available in the Playground and through the REST API.