model.py contains the model specification for passage selector and answering model. train.py can be used to train the passage selector
python3.6 -m torch.distributed.launch --nproc_per_node=4 train.py --output_dir <output_dir> dataset_cache <cache_path>
More configurations are available in t5_config.py
This code has been developed on
python3.6 transformers==2.9.1 pytorch-ignite==0.2.0