batch_size: 4 class_identifier: instruction_ranking_metric dropout: 0.1 encoder_learning_rate: 1.0e-06 encoder_model: XLM-RoBERTa keep_embeddings_frozen: true layer: mix layer_norm: false layer_transformation: sparsemax layerwise_decay: 0.95 learning_rate: 1.5e-05 load_pretrained_weights: true loss: triplet-margin nr_frozen_epochs: 0.3 optimizer: AdamW pool: avg pretrained_model: xlm-roberta-large train_data: - data/APE_train.csv validation_data: - data/APE_valid.csv warmup_steps: 0