Training
Learn how trainings work in WriftAI
Training is in Preview
Training is currently available to a limited set of users.
If you're interested in enabling training for your workspace, contact us.
What Is a Training?
Trainings let you evolve an existing model using new data or configuration. For example, you might teach a language model about your product catalog, or specialize an image model to generate visuals in your brand’s style. Instead of returning an inference result, a training run produces a new version of that model.
Trainings are executed asynchronously and follow the exact same lifecycle, webhook events, and API patterns as predictions. If you understand predictions, you already understand trainings — the only real difference is the outcome.
What a Training Involves
Inputs
Data, hyperparameters, or configuration that guide how the model should change.
Execution
WriftAI loads the current model version, allocates compute, and runs the training workload.
New Version
When training succeeds, a new model version is created and pushed to the destination model specified in the request. Once created, the new version is available immediately — you can make predictions against it.