Note: Python 3.6 or higher is needed to run the demo with the following instructions
First, navigate to the directory where you want to run the demo code in your Command Line. Then, clone this repository either by running the following command or your preferred Git cloning method. Then, navigate into the repository's directory:
git clone https://github.com/arkabhat/cse490g1-final-project.git
cd cse490g1-final-project
Before we install the required libraries, you can optionally create and activate a Python virtual environment to isolate the libraries we will be installing from your system libraries:
python3 -m venv ./venv
source ./venv/bin/activate
Then, install the required libraries:
pip3 install -r requirements.txt
Now, we can run the demo code, located in demo.py. The demo can be run by the following:
python3 demo.py <LANGUAGE-SPECIFIER>
The four valid language specifiers are ta (Tamil), ml (Malayalam), bn (Bengali) and hi (Hindi). For example, if you want to transliterate to Tamil, run the following:
python3 demo.py ta
First, the appropriate language model is loaded. Once the model is loaded, the demo prompts you for a single word (English characters only) to transliterate. You can keep providing words to be transliterated. To exit the demo, enter q.
Run the following to download the dataset tarball and extract the files:
wget -c https://storage.googleapis.com/gresearch/dakshina/dakshina_dataset_v1.0.tar
tar -xvf dakshina_dataset_v1.0.tar
rm -rf dakshina_dataset_v1.0.tar
Once the dataset has been downloaded, optionally set up and activate a Python virtual environment to isolate any libraries needed for this project from any system libraries:
python3 -m venv ./venv
source ./venv/bin/activate
Then, install all of the required libraries:
pip3 install -r requirements.txt
Jupyter should have been installed on your machine when installing the required libraries as it is one of the dependencies listed in requirements.txt.
Since attu does not have a GUI, we want to start our Jupyter notebook server without a browser. We can specify any port of our choice, and for this project, we have chosen port 8080:
# on attu
jupyter notebook --no-browser --port=8080
Once this command is run, you will notice some output from Jupyter. Specifically look for the output lines that look similar to the following:
[I 23:56:01.097 NotebookApp] Jupyter Notebook 6.4.6 is running at:
[I 23:56:01.097 NotebookApp] http://localhost:8080/?token=8b5f30b92cce57dbff42a2ef9670dcf82f79add3350c2a69
The token in the above output will be needed later.
At this point, the Jupyter notebook server is running on attu. Then, the next step is to access the Jupyter notebook server on attu through our local machine. For this, we set up an SSH tunnel from our local machine to attu (use another terminal window for this as we want our attu terminal to stay active):
# on local machine
ssh -N -L 8080:localhost:8080 <YOUR-CSE-NETID>@attu.cs.washington.edu
This command creates a tunnel from port 8080 on attu to port 8080 on our local machine. Now, navigating to localhost:8080 on our local machine should open a Jupyter window. This window will ask for a password or token to login since we are accessing the server remotely. Copy-paste the token from the Jupyter output that we noted before. Then, we will be able to access our Jupyter server on attu through our local machine and edit our notebooks.
Note: the above instructions for setting up Jupyter on attu over SSH are based on the instructions from here.