IN TRAINING : Due to communication error when running the valid dataset on a single main GPU, a multi machine and multi GPUS solution is temporarily used to run all train and valid datasets ATTENTION : Please ensure that the torch family bucket you need matches your CUDA version
To install, first create a conda environment from environment.yaml and then run pip install -r requirements.txt