deterministic: False
activations: Tanh
batch_size: 2
class_identifier: regression_metric_multi_ref
dropout: 0.5
encoder_learning_rate: 1.0e-05
encoder_model: RoBERTa
final_activation: null
hidden_sizes:
- 384
keep_embeddings_frozen: true
layer: mix
layerwise_decay: 0.95
learning_rate: 3.1e-05
load_weights_from_checkpoint: null
nr_frozen_epochs: 0.3
optimizer: AdamW
pool: avg
pretrained_model: roberta-large
topk: 3
train_data: /srv/share5/mmaddela3/metrics_project/adaptive_metrics/data/simplification/annotations_24_systems/dataset/train_grouped_refs.csv
validation_data: /srv/share5/mmaddela3/metrics_project/adaptive_metrics/data/simplification/annotations_24_systems/dataset/valid_grouped_refs.csv