LEPISZCZE-aspectemo-allegro__herbert-base-cased-v1

Description

Finetuned allegro/herbert-base-cased model on clarin-pl/aspectemo dataset.

Trained via clarin-pl-embeddings library, included in LEPISZCZE benchmark.

Results on clarin-pl/aspectemo

	accuracy	f1_macro	f1_micro	f1_weighted	recall_macro	recall_micro	recall_weighted	precision_macro	precision_micro	precision_weighted
value	0.952	0.368	0.585	0.586	0.371	0.566	0.566	0.392	0.606	0.617

Metrics per class

	precision	recall	f1	support
a_amb	0.2	0.033	0.057	91
a_minus_m	0.632	0.542	0.584	1033
a_minus_s	0.156	0.209	0.178	67
a_plus_m	0.781	0.694	0.735	1015
a_plus_s	0.153	0.22	0.18	41
a_zero	0.431	0.529	0.475	501

Finetuning hyperparameters

Hyperparameter Name	Value
use_scheduler	True
optimizer	AdamW
warmup_steps	25
learning_rate	0.0005
adam_epsilon	1e-05
weight_decay	0
finetune_last_n_layers	4
classifier_dropout	0.2
max_seq_length	512
batch_size	64
max_epochs	20
early_stopping_monitor	val/Loss
early_stopping_mode	min
early_stopping_patience	3

Citation (BibTeX)

@article{augustyniak2022way,
  title={This is the way: designing and compiling LEPISZCZE, a comprehensive NLP benchmark for Polish},
  author={Augustyniak, Lukasz and Tagowski, Kamil and Sawczyn, Albert and Janiak, Denis and Bartusiak, Roman and Szymczak, Adrian and Janz, Arkadiusz and Szyma{'n}ski, Piotr and W{\k{a}}troba, Marcin and Morzy, Miko{\l}aj and others},
  journal={Advances in Neural Information Processing Systems},
  volume={35},
  pages={21805--21818},
  year={2022}
}