dpr/configs/training.yaml

task: dpr
base_training_updates: 4000
base_training_processes: 8
region_refinement_required: true
feature_reconstruction_required: true