Build Documentation

Date: 30.05.2023

Note: Windows is currently no longer supported since the new vector database weaviate is used (18.07.23)

This is a textual writedown of the information of following video: Deployment Video and some additional information.

Local Deployment

1. Clone the Github Repo:

git clone https://github.com/amosproj/amos2023ss03-qachat

2. Set Python Path:

Go to the cloned repository and there set the Python-Path. This will look something like:

Windows:

$env:PYTHONPATH += "PATH_TO_THE_REPOSITORY"

Unix:

export PYTHONPATH=$PYTHONPATH:PATH_TO_THE_REPOSITORY

3. Create python environment, activate it & load required dependencies

First create the Virtual Python Enviroment:

Windows:

python.exe -m venv venv

Unix:

python -m venv venv

---

After this activate it with:

Windows:

.\venv\Scripts\activate

Unix:

./venv/bin/activate

---

Now you activated the environment and need to download the required dependencies:

Windows:

pip install -r .\requirements.txt

Unix:

pip install -r ./requirements.txt

When running the DataProcessing there can be a problem, that it can not find the needed spacy packages. It will be something like can not find package de_name_name_sm. To fix this download the needed models. Therefore after installing the requirements run the following commands:

python -m spacy download xx_ent_wiki_sm
python -m spacy download de_core_news_sm

4. Add the tokens to the project

For the project you need a variety of tokens, to get the project running. These are 3 Slack, 1 DeepL, 1 Tessdata and 3 Confluence Tokens, 2 Google Cloud (when you want to host the server on google cloud). Add them by creating a file called 'tokens.env' in the 'QAChat' Folder. The tokens you either need to get from the services themselves, as described in the associated Wiki Pages, or get them from the secure channel.
The exception is the Tessdata "token", which is simply the path to your tessdata directory. The directory can be found within the virtual environment you created for this project.

5. (Optional) - Testing

For starting the implemented test first switch to the 'Testing' folder. Now you can run:

pytest

6. (Optional) - Loading Data into the Weaviate Database

This will be automated in the end and after the initial setup, won't be necessary. For completeness of this guide, this still should be mentioned. First make sure you installed poppler. This is needed, if pdf files need to be analyzed as pictures. Therefore on a Mac with homebrew one can run:

brew install poppler

On other you will need to install it manually. Please look this up.

After that, change into the 'QAChat\DataProcessing' folder and there run the main.py file by entering following into your terminal:

Windows:

python.exe .\main.py

Unix:

python ./main.py

This can take quite some time, as there are already Machine Learning Models being used here. This will take even more time on the first time you run the program, as some models may need to be downloaded first.

When the process is finished, the database is filled with data, that can then be used for answering the question of users.

7. Starting the Slack-Bot

To start the Slack Bot change to the 'QA_Chat\Slack_Bot' folder.

Then run:

Windows:

python.exe .\qa_agent.py

Unix:

python ./qa_agent.py

As soon as you see the message: 'Bolt app is running!' in your terminal, the Slack Bot will react to any message it gets sent on Slack in its message tab. On how to use the Slack Bot on Slack, take a look at the User Documentation

7. Run the QA Bot

To run the qa bot run: Windows:

python.exe .\setup_server.py

Unix:

python ./setup_server.py

in the QAChat/QA_Bot folder.

When you want to use GPU acceleration please make sure CUDA is installed and run:

CMAKE_ARGS="-DLLAMA_CUBLAS=on"  pip install llama-cpp-python --force-reinstall --upgrade --no-cache-dir