In this project, I trained some well-known architectures for generative question answering and evaluated these models on ROUGE-L metric. I also created a database to provide supporting documents for these model when inferring.

pip install -r requirements

On Ubuntu:

# Create the file repository configuration:
sudo sh -c 'echo "deb http://apt.postgresql.org/pub/repos/apt $(lsb_release -cs)-pgdg main" > /etc/apt/sources.list.d/pgdg.list'

# Import the repository signing key:
wget --quiet -O - https://www.postgresql.org/media/keys/ACCC4CF8.asc | sudo apt-key add -

# Update the package lists:
sudo apt-get update

# Install PostgreSQL and supporting libraries.
# In this project, I'm using version 12, so I specify the installation as follow
sudo apt-get -y install postgresql-12 postgresql-server-dev-12 libpq-dev

Before inference or inserting knowledge to the database, remember to start the Postgresql first by using these commands:

# Check if Postgresql has started yet 
sudo service postgresql status

# If the status is 'down', we have turn it on
sudo service postgresql start

pgvector is a powerful tool that helps storing vectors in Postgresql databases and searching vectors computing similarity with the given query. pqvector supports methods such as:

• Exact and approximate nearest neighbor search
• L2 distance, inner product, and cosine distance

To install on Ubuntu, please see this installation guide. If you run into problems with make command, this discussion may help.

To train a model defined in this project, you have to set up parameters for training in a config file. You can find the sample of this config file in configs directory. Finally, you pass the path to the config file to this command to start training.

python3 train.py -c <path-to-config-file>

For example:

python3 train.py -c configs/t5/t5_small.json

Before inferring, remember to start Postgresql service first (see Postgresql section). Note that, configurations for inference must also be set up in a config file and its path has to be passed down to this command. You can find the sample of the file here.

python3 infer.py

Before interacting with the knowledge hub, remember to start Postgresql service first (see Postgresql section). Note that, configurations for knowledge hub must also be set up in a config file and its path has to be passed down to this command. You can find the sample of the file here.

python3 knowledge_hub.py -n <number-of-samples-to-be-inserted>

For example:

# Inserting 1,000,000 samples
python3 knowledge_hub.py -n 1000000

To see how I use pgvector and Postgresql to store and search vectors by using some common methods to compute similarity, checkout pqvector guide and here.