Marv

• 

  Dialogue1

• 

  Dialogue2

Inspiration

I wanted to implement a GPT model into Unity

What it does

Marv uses GPT-neoX-20B to predict and respond to anything like a human being. Essentially a GPT model uses statistics to predict an output based on a specific input. The GPT-neoX-20B model has a very large amount of parameters, thus is able to recognize, understand and produce text that sounds human.

How I built it

Unfortunately Unity does not support Python and the model API can only be implemented in Python. So, I implemented the api in a python code, then I used sockets in a C# code to communicate between the 2 codes simultaneously and then I was able to use the model from Unity.

Challenges I ran into

The hardest part was training and tuning the model to make it sound like a normal human being. I also ran into other problems like getting blank outputs from the model.

Accomplishments that I'm proud of

This is my first attempt implementing a language model for NLP into Unity.

What I learned

I learnt how to communicate between different programming languages.

What's next for Marv

I'm planning to add more NPC's and improve the accuracy of the model.

Built With

• c#
• gpt
• python
• sockets
• unity