seed: 42
monitor: pearson
metric_mode: max
early_stopping: True
patience: 1
min_delta: 0.0
save_top_k: 2
save_weights_only: False
min_epochs: 1
max_epochs: 100
gradient_clip_val: 1.0
gpus: 1
precision: 32

batch_size: 64
accumulate_grad_batches: 4
loader_workers: 4

optimizer: Adam
learning_rate: 3.0e-05
encoder_learning_rate: 1.0e-05
layerwise_decay: 0.95
nr_frozen_epochs: 100000
scheduler: constant

train_path: data_en/polaris/polaris_train.csv
val_path: data_en/polaris/polaris_val.csv
test_path: data_en/polaris/polaris_test.csv
train_img_dir_path: data_en/polaris/images
val_img_dir_path: data_en/polaris/images
test_img_dir_path: data_en/polaris/images

model: PolosEstimator
loss: mse
encoder_model: BERT
# pretrained_model: princeton-nlp/sup-simcse-roberta-large
pretrained_model: princeton-nlp/sup-simcse-roberta-base

layer: mix
scalar_mix_dropout: 0.1
pool: cls

dropout: 0.1
activations: Tanh
hidden_sizes: "2304,1152"
final_activation: False