Code for home project of IN9550 UiO.

Run the following command to train a tokenizer.

python -m nmt.tokenizer --vocab_size <vocabulary size>

The training configuration is defined in a YAML file and can be overrided using command line arguments. Each time the command is run the configuration is saved under data/runs/{model_name}.meta.yaml

python -m nmt.main -c config/transformer_default_config.yaml

python -m nmt.main -c config/rnn_default_config.yaml

The configuration file used to run evaluation is a YAML file with the following entries. An example is available here.

• tokenizer: Path to the tokenizer
• test_data: Path to test dataset
• model: Path to Transformer-based Seq2Seq model
• output_file: Path to save predictions
• debug: Whether or not to run in debug mode

Run the following command to evaluate your Transformer-based model:

python -m nmt.evaluation.transformer_evaluation -c <config_file>

Run the following command to evaluate your RNN-based model:

python -m nmt.evaluation.rnn_evaluation -c <config_file>

The code in this repository is mainly based on Pytorch tutorials.