npc-engine

NPC-Engine is a deep learning and NLP toolkit for designing NPC AI with natural language.

Features

• Chat-bot dialogue system.
• SoTA tools like text semantic similarity and text to speech.
• Easy, open source deep learning model standard (ONNX with YAML configs).
• GPU accelerated inference with onnxruntime.
• Engine agnostic API through ZMQ server via JSONRPC 2.0.

Getting started

The easiest way to get started is to use NPC Engine through the Unity integration

You can also use it directly through ZMQ or HTTP. See Documentation for more details.

Roadmap

Done:

• Real-time end-to-end chatbot dialogue system
• Semantic similarity
• Real-time speech to text (experimental)
• Unity integration
• CLI tool for importing models from Huggingface
• Asynchronous API features

In progress:

• Actions and planning
• Unreal integration
• Importing models from popular TTS libraries
• Emotion features
• Multiple languages support
• Much more

Build on Windows

• Create virtualenv and activate it:

  > python3 -m venv npc-engine-venv
  > .\npc-engine-venv\activate.bat

• Install dependencies

  > pip install -e .[dev,dml]
  

• (Optional) Compile, build and install your custom ONNX python runtime

  Build instructions here https://onnxruntime.ai/

• (Optional) Run tests

  > tox
  

• Compile to exe with:

  > pyinstaller --hidden-import="sklearn.utils._cython_blas" --hidden-import="sklearn.neighbors.typedefs" ^
  --hidden-import="sklearn.neighbors.quad_tree" --hidden-import="sklearn.tree._utils" ^
  --hidden-import="sklearn.neighbors._typedefs" --hidden-import="sklearn.utils._typedefs" ^
  --hidden-import="sklearn.neighbors._partition_nodes" --additional-hooks-dir hooks ^
  --exclude-module tkinter --exclude-module matplotlib .\npc_engine\cli.py --onedir
  

Docker

If you wish to host NPC Engine somewhere you can use our the docker image. It's Linux image with either TensorRT or CPU ONNX Runtime providers.

You can build it yourself with:

docker build -t npc-engine . -f Dockerfile.cuda

or

docker build -t npc-engine . -f Dockerfile.cpu

To run the image you must mount the models directory to /app/models e.g.

docker run --gpus all -it --mount type=bind,source=%cd%\tests\resources\models,target=/app/models -p 5000:5000 npc-engine/inference-engine:latest npc-engine run --port 5000

Where --gpus all will give access to the GPU (not needed when built for CPU), -it will output logs and let you use the container interactively, --mount will mount the models directory to the container, -p 5000:5000 will expose the port 5000 on the host machine.

Community

We have a Discord server where you can get support, ask questions and show off your creations.

If you would like to donate, you can check out our Patreon.

Our Patrons

• Marrech Games
• עינ הרע

Authors

• eublefar - Python, Neural Nets - github
• igorzmitrovich - Python, CI/CD - github

See also the list of contributors who participated in this project.