An AI assistant for jMonkeyEngine and related projects.

It knows :

• all the main documentation of jMonkeyEngine and some prominent projects
• the entire source code of the jMonkeyEngine main repo
• how to search the forum for basic queries

For more details, check the knowledge base section.

To provide the functionality of the bot, the following libraries are used:

• 🦜🔗 LangChain: to interface with GPT-3
• 🤗 HuggingFace: inference libraries and models
• OpenAI apis: LLM for text generation
• MiniLM L6 v2: fast embeddings
• all-mpnet-base-v2: embeddings
• Flan-t5-base-samsum: Summarization
• Sumy: Alternative summarization
• Yake: keywords extraction
• LibreTranslate: Multi language support and detection
• and more .. Check the requirements.txt and environment.yml for the full list.

The bot extends the knowledge of GPT-3 by embedding pieces of information from the following sources:

Static embeddings are updated periodically and stored in the embeddings/ folder in this repo.

• jmonkeyengine.org (partial)
• jMonkeyEngine Wiki
• jMonkeyEngine source code
• Minie Wiki
• Minie source code
• Lemur Wiki
• Lemur source code
• Zay-ES wiki
• Zay-ES source code
• Monkey-Droid/jme3-Simple-Examples

Dynamic embeddings are generated on the fly for the requested information.

• jMonkeyEngine Hub (simple search queries based on keywords, cut off: 2023-02-03 )

1. Install Miniconda or Anaconda
2. Clone this repo and cd into it
3. Create the environment

conda env create  -f environment.yml

4. Activate the environment

conda activate jmebot

5. Install extra dependencies

pip install -r requirements.txt

5b. If you want to run it on a GPU and you have a CUDA compatible GPU, make sure to have a recent version of CUDA installed and then install the required dependencies

# For Ubuntu
sudo apt-get install  build-essential cmake swig libopenblas-dev libpcre3-dev
bash installForCuda.sh # Install the required dependencies (note: this builds faiss-gpu from source, so it will take a while)

6. Export your OpenAI API key

export OPENAI_API_KEY="XXXXX"

7. Tweak the config.json file, if you want to run it on a CPU you should change DEVICE to "cpu" and USE_SUMY to true)
8. Run the bot

bash bot.sh

or regenerate the embeddings

bash bot.sh ingest

In a docker host

The snippets below show how to use the prebuild images on github registry, If you want to build your own image:

# For cpu
docker build -t chat-jme .

# For cuda
docker build -t chat-jme:cuda . -f Dockerfile.cuda

mkdir -p /srv/chat-jme/cache
chown -Rf 1000:1000 /srv/chat-jme/cache 

# For CPU
docker run -d --restart=always \
-eOPENAI_API_KEY="XXXXXXXX" \
-v/srv/chat-jme/cache:/home/nonroot/.cache \
-p8080:8080 \
--name="chat-jme" \
ghcr.io/riccardobl/chat-jme/chat-jme:snapshot bot

# For Cuda (recommended)
GPUID="device=GPU-XXXXX"
docker run -d --restart=always \
-eOPENAI_API_KEY="XXXXXXXX" \
-v/srv/chat-jme/cache:/home/nonroot/.cache \
-p8080:8080 \
--gpus $GPUID
--name="chat-jme" \
 ghcr.io/riccardobl/chat-jme/chat-jme:cuda-snapshot bot

NOTE: To use custom static embeddings specify the INDEX_PATH environment variable

NOTE2: the first run might take some time since it has to download the models.

NOTE3: If you use the cpu you might need to add --security-opt seccomp=unconfined to the docker command if performances are bad (note that this is not recommended)

mkdir -p /srv/chat-jme/cache
chown -Rf 1000:1000 /srv/chat-jme/cache 

mkdir -p /srv/chat-jme/embeddings
chown -Rf 1000:1000 /srv/chat-jme/embeddings

docker run -d --restart=always \
-eOPENAI_API_KEY="XXXXXXXX" \
-eINDEX_PATH="/embeddings" \
-v/srv/chat-jme/cache:/home/nonroot/.cache \
-v/srv/chat-jme/embeddings:/embeddings \
--name="chat-jme" \
ghcr.io/riccardobl/chat-jme/chat-jme:snapshot ingest

POST /session

REQUEST

{
    "sessionSecret":"", // sessionSecret of the session to maintain or nothing to create a new one
    "lang":"en" // || "it" || etc... || "auto",    
}

RESPONSE

{
    "sessionSecret":"XYZ", // sessionSecret of the session
    "helloText":"???", // Text that can be used to initiate a conversation with the bot (in the chosen language)
    "welcomeText": "..." // Hardcoded welcome text in the specified language
}

POST /query

REQUEST

{
    "sessionSecret":"",
    "lang":"en",// || "it" || etc... || "auto",
    "question":"Your question"
}

RESPONSE

{
    "output_text":"???" // Answer to the question
}

GET /lang

RESPONSE

[
    {
        "name":"English",
        "code":"en"
    },
    {
        "name":"Italian",
        "code":"it"
    },
    ...
]

The frontend is server on the 8080 port by default. It supports some configuration parameters that can be passed as document hash parameters. Multiple parameters can be concatenated with the & character.

• lang the language for questions and answers. (default: en, use auto to get the bot to detect the language automatically)
• compact if true, the bot will be displayed in a compact mode, good for embedding (default: false)