Inference
An inference endpoint serves an open-weight model over an OpenAI-compatible HTTP API. You pick a model from the catalog, or a checkpoint you fine-tuned, and the platform deploys it on GPUs and gives you a URL. Inference endpoints, datasets and fine-tuning jobs together make up Token Factory, the part of Saturn Cloud that handles the open-model lifecycle.
Anyone can sign up and create an inference endpoint to try the platform. Production endpoints run on dedicated GPU capacity that we arrange for each customer. Talk to us about dedicated GPU capacity.
Start here
| Page | What it covers |
|---|---|
| Quickstart | Deploy a model and send it a request |
| Calling an Endpoint | Endpoint URLs, authentication, model names and example requests |
| Inference Endpoints | Every option when creating an endpoint, and how to manage it |
Fine-tuning
You can fine-tune an open base model on your own data and serve the result. A dataset feeds a fine-tuning job, the job produces a checkpoint, and the checkpoint is served by an inference endpoint. Several fine-tuned adapters can share one dedicated endpoint for the same base model.
| Page | What it covers |
|---|---|
| Concepts | Datasets, jobs, checkpoints and endpoints, and how they relate |
| Datasets | Registering and sealing training data |
| Fine-Tuning Jobs | Starting and monitoring a fine-tuning job |
Token Factory objects belong to the organization, not to one user. A dataset you register is visible to everyone in your organization, and a checkpoint a teammate produces is one you can serve.