activations: Tanh
batch_size: 16
class_identifier: referenceless_regression_metric
dropout: 0.2
encoder_learning_rate: 4.697617286722686e-05
encoder_model: XLM-RoBERTa
final_activation: null
hidden_sizes:
- 2048
- 1024
keep_embeddings_frozen: true
layer: mix
layer_norm: false
layer_transformation: sparsemax
layerwise_decay: 0.95
learning_rate: 0.0002369203764556797
load_pretrained_weights: true
local_files_only: false
loss: mse
nr_frozen_epochs: 0.3
optimizer: AdamW
pool: avg
pretrained_model: FacebookAI/xlm-roberta-large
train_data:
- data/data-42-last/train.csv
validation_data:
- data/data-42-last/test.csv
warmup_steps: 0