Tabnine's fine-tuned AI models

Tabnine’s fine-tuned AI models (Tabnine Enterprise only)

In some cases, Tabnine offers Enterprise customers the ability to deploy a fine-tuned AI model instead of the universal code model. The fine-tuned AI model is a refined Universal model that’s essentially retrained on the customer’s private code. Fine-tuned models are trained in and hosted on the same deployment option (per the customer’s choice).

Fine-tuning AI models

To train a fine-tuned AI model, the latest version of Tabnine’s Universal model is cloned into the Tabnine Trainbox. Using the relevant Tabnine codebase integration, the customer selects the relevant repositories from their codebase. Tabnine’s Trainbox then retrains the cloned Universal model with the private customer code, resulting in a refined version of the Universal model.

This fine-tuned AI model is then deployed to the customer setup (instead of the Universal model). The code completions in this setup are served from the fine-tuned AI model. Only the users in this enterprise account get code completion from this model.

The fine-tuned AI model is retrained and updated according to the customer's desired frequency.