activations: Tanh
additional_score_in:
- false
- false
- false
- false
- false
additional_score_out:
- false
- false
- false
- false
- false
additional_translation_in:
- false
- false
- false
- false
- false
backprop_additional_translation: true
batch_size: 16
class_identifier: polycand_metric
dropout: 0.1
encoder_learning_rate: 1.0e-06
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: 1.5e-05
load_pretrained_weights: true
local_files_only: false
loss: mse
nr_frozen_epochs: 0.3
optimizer: AdamW
pool: avg
pretrained_model: xlm-roberta-large
train_data:
- data/csv/train_same_rand_wmt25.csv
use_ref: false
validation_data:
- data/csv/dev_same_rand.csv
warmup_steps: 0