indobert-fin_news-mlm-3

This model is a fine-tuned version of indobenchmark/indobert-base-p1 on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 2.1666

Model description

This model is an experiment based on the methods in the paper arXiv:2310.09736 [cs.CL] by post-training the model indobert-base-p1 on the Financial News Articles dataset. The dataset was pre-processed as follows:

  1. Cleaning the dataset (mainly removing article header/location info, "Baca Juga" occurrences, and standard footers about Google News).
  2. Segmenting the dataset using the model sat-3l-sm from wtpsplit.

The post-training arguments follow that of the paper. Post-training takes a little over 8 hours (2x T4 GPU).

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

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: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
  • lr_scheduler_type: linear
  • num_epochs: 50
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss
4.3212 1.0 740 4.1454
3.7192 2.0 1480 3.5316
3.3165 3.0 2220 3.2571
3.0983 4.0 2960 3.0149
2.8676 5.0 3700 2.9255
2.811 6.0 4440 2.8177
2.7158 7.0 5180 2.7367
2.5976 8.0 5920 2.7042
2.4428 9.0 6660 2.5891
2.3634 10.0 7400 2.5273
2.3411 11.0 8140 2.4821
2.2963 12.0 8880 2.4501
2.2155 13.0 9620 2.3981
2.0997 14.0 10360 2.4257
2.1241 15.0 11100 2.3665
2.0828 16.0 11840 2.3150
1.9577 17.0 12580 2.3608
1.9807 18.0 13320 2.2961
1.9717 19.0 14060 2.2521
1.8938 20.0 14800 2.2366
1.8353 21.0 15540 2.2804
1.905 22.0 16280 2.2110
1.8405 23.0 17020 2.1378
1.8043 24.0 17760 2.2252
1.7745 25.0 18500 2.1675
1.6882 26.0 19240 2.1594
1.6651 27.0 19980 2.1635
1.672 28.0 20720 2.1666

Framework versions

  • Transformers 4.51.3
  • Pytorch 2.6.0+cu124
  • Datasets 3.6.0
  • Tokenizers 0.21.1
Downloads last month
5
Safetensors
Model size
0.1B params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for elidle/indobert-fin_news-mlm

Finetuned
(103)
this model
Finetunes
5 models

Evaluation results