IndoBERT Lite Base IndoNLI Distil mDeBERTa

IndoBERT Lite Base IndoNLI Distil mDeBERTa is a natural language inference (NLI) model based on the ALBERT model. The model was originally the pre-trained indobenchmark/indobert-lite-base-p1 model, which is then fine-tuned on IndoNLI's dataset consisting of Indonesian Wikipedia, news, and Web articles [1], whilst being distilled from MoritzLaurer/mDeBERTa-v3-base-xnli-multilingual-nli-2mil7.

Evaluation Results

	`dev` Acc.	`test_lay` Acc.	`test_expert` Acc.
`IndoNLI`	77.19	74.42	61.22

Model

Model	#params	Arch.	Training/Validation data (text)
`indobert-lite-base-p1-indonli-distil-mdeberta`	11.7M	ALBERT Base	`IndoNLI`

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 16
eval_batch_size: 16
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.5053	1.0	646	0.4511	0.7506	0.7462	0.7530	0.7445
0.4516	2.0	1292	0.4458	0.7692	0.7683	0.7684	0.7697
0.4192	3.0	1938	0.4433	0.7701	0.7677	0.7685	0.7673
0.3647	4.0	2584	0.4497	0.7720	0.7699	0.7697	0.7701
0.3502	5.0	3230	0.4530	0.7679	0.7661	0.7658	0.7668

Framework versions

Transformers 4.36.2
Pytorch 2.1.2+cu121
Datasets 2.16.1
Tokenizers 0.15.0

References

[1] Mahendra, R., Aji, A. F., Louvan, S., Rahman, F., & Vania, C. (2021, November). IndoNLI: A Natural Language Inference Dataset for Indonesian. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics.