# seed 8
activations: Tanh
batch_size: 4
class_identifier: referenceless_regression_metric
dropout: 0.15
encoder_learning_rate: 1.0e-05
encoder_model: XLM-RoBERTa
hidden_sizes:
- 2048
- 1024
keep_embeddings_frozen: true
layer: mix
layerwise_decay: 0.95
learning_rate: 1.0e-05
load_weights_from_checkpoint: null
nr_frozen_epochs: 1
optimizer: AdamW
pool: avg
pretrained_model: xlm-roberta-large
train_data: data/mqm.train.z_score.csv
validation_data: data/mqm.test.z_score.csv