This is the in-class activity for Tuesday January 25.
The materials that you will need for this in-class activity are:
In Class Activity: Intent Determination with GPT-3
Today in class you will use the data you created for intent determination in text adventure games to train a GPT-3 model to perform the intent determination.
In our first version of the Text Adventure Game, we had a simple keyword-based method of handling the player’s intent. The parser implemented a function called get_player_intent that took in a player’s command and then decided which of the following actions a player was trying to accomplish:
- Move to a new location
- Examine an object
- Get an object
- Drop an object
- Perform a special action using an object
- Check the player’s inventory
- Redescribe the scene
- Perform a sequence of actions
The Colab Notebook will guide you through how to train a model to perform intent detection using the data we collected
What to do
- Open the GPT3 for Intent Determination.ipynb.
- Choose
Runtime > Run all. - The Colab Notebook will ask “Permit this notebook to access your Google Drive files?”. You should grant it permission.
- You may have to get your Intent annotation data from last time from Gradescope
- Scroll down to the “START HERE: Create a Prompt for Intent Detection” section of the notebook
- Complete the interactive annotation activity. It will save a JSON file to your Google drive.
- Please submit your work to Gradescope by Monday, January 31, 2022 before 11:59PM.