Reinforcement Learning Training

Now, that we have defined a reward function, we are ready to start a fine-tuning job on the Augento Platform.

In the sidebar, go to Training and select the Start Training tab.

Select the model, which you previously connected to your agent in the dropdown menu.

Select Training Data

Now, in the table view, select the queries you want your fine-tuned model to be trained on. In most cases, these are queries where your running production system failed and where you want the model to improve on.

Connect Reward Function

Now connect the reward function. Click on Connect Reward Function and simply paste the URL of the endpoint from the previous step.

Training parameters

You can change the hyperparameters of the training, if you wish. However, we already provide defaults defaults that will fit most use cases.

Start Training

No you can start the fine-tuning training by clicking Start Training. You can view the submitted job in the Training Runs tab on the same page.