indobert-lite-base-p2

This model is a fine-tuned version of indobenchmark/indobert-lite-base-p2 on the indonlu-smsa and id_google_play_review dataset. It achieves the following results on the evaluation set for combined dataset from indonlu-smsa and id_google_play_review:

• Loss: 0.4257
• Accuracy: 0.9291
• Precision: 0.8637
• Recall: 0.8651
• F1: 0.8643

Seperate evaluation indonlu/indonlu-smsa

• Accuracy: 0.9269
• Precision: 0.9067
• Recall: 0.8948
• F1: 0.89995

Model description

https://huggingface.co/indobenchmark/indobert-lite-base-p2

To Do:

• Add optimized model from optimum

Intended uses & limitations

Sentiment Analysis, this model more lightweight than bert base and roberta base ofc because this is lite model haha

Training and evaluation data

The training combined all training data from indonlu-smsa and id google play review The evaluation is conducted using the validation split

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-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
• lr_scheduler_warmup_steps: 500
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
0.0744	1.0	1127	0.3146	0.9309	0.8846	0.8549	0.8691
0.056	2.0	2254	0.3321	0.9298	0.8737	0.8580	0.8650
0.0299	3.0	3381	0.3906	0.9316	0.8750	0.8656	0.8702
0.0317	4.0	4508	0.4257	0.9291	0.8637	0.8651	0.8643

Framework versions

• Transformers 4.48.2
• Pytorch 2.6.0+cu126
• Datasets 3.2.0
• Tokenizers 0.21.0