batch_size: 4
context: true
context_turns: 3
dataset: dailydialog
dataset_path: data/dailydialog/
encoder_learning_rate: 1.0e-05
labels: dailydialog
layerwise_decay: 0.95
learning_rate: 5.0e-05
nr_frozen_epochs: 1
pretrained_model: roberta-base