nr_frozen_epochs: 0.3
keep_embeddings_frozen: true
optimizer: AdamW
encoder_learning_rate: 1.0e-05
learning_rate: 3.1e-05
layerwise_decay: 0.95
encoder_model: XLM-RoBERTa
pretrained_model: xlm-roberta-large
pool: avg
layer: mix
dropout: 0.15
batch_size: 8
class_identifier: referenceless_regression_metric
train_data: data/jan_28_ende_num_0.5_del_None_mask_None_xlm_mbart.csv
validation_data: None
hidden_sizes:
- 2048
- 1024
activations: Tanh
final_activation: null
load_weights_from_checkpoint: null