Fine-tuning Azure OpenAI models in Azure AI Foundry

Now you are ready to start training your fine-tuned model. This is a batch process, and since it requires significant resources, your job may sit in the queue for some time. Once accepted, the run can take several hours, especially if you are working with a large, complex model and large training data set. Azure AI Foundry tools allow you to see the status of a fine-tuning job, showing results, events, and hyperparameters used.

Each pass through the training data creates a checkpoint. This is a useful version of the model in its current state of tuning, so you can evaluate it with your code before finishing your fine-tuning work. You’ll always have access to the latest three outputs, so you can compare different versions before making your final choice.

Ensuring the safety of finely tuned models

Microsoft’s own AI safety rules apply to your finely tuned model. It will not be published until you explicitly choose to publish it, with testing and evaluation in private workspaces. At the same time, your training data remains private and is not stored with the model, reducing the risk of sensitive data being leaked through operational attacks. Microsoft will scan training data before it is used to ensure it does not contain malicious content, and will abort the job before it runs if it detects inappropriate content.