To fine tune an OpenAI model, you have to first select a model which is eligible for Fine Tuning (currently : Ada, Babbage, Curie, DaVinci).

We recommend the DaVinci model as it provides the greatest performances.

You will then be prompted to upload a file containing your fine tune training :

Capture d’écran 2023-03-31 à 17.31.47.png

The requested file is a .csv file that you can set up in a spreadsheet application like Microsoft Excel, Google Sheets or Apple Numbers.

The training format consists in a simple table of two columns :

Capture d’écran 2023-03-31 à 17.32.34.png

The first column represents examples of queries that your users may submit to your bot, and the second one, the corresponding responses.

<aside> ✅ Don't hesitate to provide variations of examples. Keep in mind that the more, the better !

</aside>

Once your are done with your spreadsheet, just export it to a comma-sperated csv and upload it in your bot. Launch the training, and you are good to go !

<aside> ⚠️ Due to OpenAi's current success, training processes may take a while so we recommand that you request the fewest trainings possible.

</aside>