QGAR - Question Generation Answering Responder

Welcome to the QGAR Project - a note-processing and flashcard generation application.

If you want to run the application, please read the Prerequisites section.

Table of Contents

• 1. Prerequisites
  • 1.1 Create Access Token Files
  • 1.2 Install Dependencies
• 2. Run QGAR
• 3. Setup Virtual Environment

---

1. Prerequisites

1.1 Create Access Token Files

To get access to our models and datasets you must be a part of our organization, The Coorporation, on Huggingface Hub.

Moreover, when training the QG model, the training process will be uploaded to wandb.

In order for the QGAR to do the above, you must be authorized via access tokens.

Therefore you must create a .local directory in the root directory and create the following files containing your tokens:

• hg_token.txt - contains your access token to Huggingface
• wandb_token.txt - contains your access token to Weight and Biases

1.2 Install Dependencies

QGAR uses 3rd party libraries which are specified in requirements.txt.

To install them either use the command make install or:

pip install -r requirements.txt

NB: Make sure you install the libraries in a Virtual Environment!

---

2. Run QGAR

You run QGAR by running main.py via:

python main.py <args>

The application can take arguments to specify what to do:

• main.py -t - starts a training session for the QG model.
• main.py -qg <context> - runs the QG model on the input text and generates questions from the given context.
• main.py -d - downloads and preprocesses our modified SQuAD dataset
• main.py -n <path> - the path to a note file
• main.py -s <filename>.json - use the specified settings file. Defaults to settings.json
• main.py -h - shows instructions on how to run the program
• main.py -o - used in combination with -n; specifies whether to output questions and answers as json (debug) or csv. Defaults to csv.

---

3. Setup Virtual Environment

The following sections describe the setup process to use Hugging Face and PyTorch locally on your device running in a Virtual Environment.

The steps have been tested to run without errors on Linux 22.04 (LTS).

Table of Contents:

• 3.1 Setup Miniconda
• 3.2 Install PyTorch with NVIDIA GPU
• 3.3 Install HuggingFace Transformers
• 3.4 Configure VS Code to use Virtual Environment
• 3.5 Setup on MAC M1

---

3.1 Setup Miniconda

1. Install Miniconda

   Miniconda will be in charge of managing our virtual Python environment and will download and install needed packages.

   To install it, run the commands:

   curl https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -o Miniconda3-latest-Linux-x86_64.sh
   
   bash Miniconda3-latest-Linux-x86_64.sh

2. Create Virtual Environment

   To isolate our development environment from the main Python installation on your system, we will create a virtual environment.

   In this, we will install all the needed Python packages, which will only be available through the environment.

   Run the following command, where <name> is the name of your local environment:

   conda create --name <name> python=3.10.9

3. Activate the Virtual Environment

   Now we must activate our created environment, so all install commands targets the virtual environment and not your main installation.

   Run the following command, where <name> is the name of the virtual environment you created in step 2.

   conda activate <name>

   In the terminal you should now see the name of your local environment. This way you always know if it is activated.

---

3.2 Install PyTorch with NVIDIA GPU

1. Activate your Virtual Environment

   First, make sure you have activated your virtual environment.

   You can activate it with the following command, where <name> is the name of the environment:

   conda activate <name>

2. Install PyTorch

   Run the following command:

   conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia

3. Verify the installation

   Run the following command:

   python3 -c "import torch; print(torch.cuda.is_available())"

   It should return True in the console!

---

---

3.3 Install HuggingFace Transformers

1. Activate your Virtual Environment

   First, make sure you have activated your virtual environment.

   You can activate it with the following command, where <name> is the name of the environment:

   conda activate <name>

2. Install Transformers

   Run the following command:

   conda install -c huggingface transformers

3. Verify installation

   Run the following command:

   python -c "from transformers import pipeline; print(pipeline('sentiment-analysis')('we love you'))"

   It should output something similar to the following:

   [{'label': 'POSITIVE', 'score': 0.9998704791069031}]

---

3.4 Configure VS Code to use Virtual Environment

1. Open example.py to activate the VS Code Python Extension.

2. Press the F1 key on your keyboard

3. Write the following:

   > Python: Select Interpreter

4. Select the interpreter with the name of your local environment (listed as conda).

5. The libraries you have installed are now imported correctly.

---

3.5 Running on Apple Silicon (M1)

Follow instructions here link

to run, remember to use

python3 [file].py

And not just "python"

To install packages, use

python3 -m pip install [package]

---