blacklotusid/id-hs-indobert

indonesian-hate-speech-bert

This model is a fine-tuned version of indobenchmark/indobert-base-p1 on https://github.com/okkyibrohim/id-multi-label-hate-speech-and-abusive-language-detection dataset. It achieves the following results on the evaluation set:

• Loss: 0.3245
• Accuracy: 0.8808
• F1: 0.8808
• Precision: 0.8808
• Recall: 0.8808
• Precision Non Hate: 0.8954
• Precision Hate: 0.8609
• Recall Non Hate: 0.8978
• Recall Hate: 0.8578

Model description

More information needed

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: 32
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• 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_ratio: 0.1
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall	Precision Non Hate	Precision Hate	Recall Non Hate	Recall Hate
0.3074	1.3618	200	0.2969	0.8817	0.8818	0.8820	0.8817	0.9020	0.8549	0.8915	0.8685
0.1442	2.7235	400	0.3146	0.8933	0.8933	0.8933	0.8933	0.9080	0.8733	0.9065	0.8753
0.0699	4.0819	600	0.4224	0.8875	0.8874	0.8874	0.8875	0.9003	0.8699	0.9048	0.8639

Framework versions

• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Tokenizers 0.21.1