indicbert-hatespeechdetection-telugu

This model is a fine-tuned version of ai4bharat/IndicBERTv2-MLM-only on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2457
• Accuracy: 0.9464
• F1: 0.9600

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: 32
• 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: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1
0.6426	1.0	14	0.5502	0.6429	0.7826
0.5289	2.0	28	0.4144	0.8393	0.8889
0.4047	3.0	42	0.2915	0.875	0.9067
0.2643	4.0	56	0.2390	0.8929	0.9211
0.1543	5.0	70	0.1751	0.9286	0.9444
0.1073	6.0	84	0.1605	0.9464	0.9589
0.0604	7.0	98	0.2654	0.8929	0.9211
0.0335	8.0	112	0.1993	0.9464	0.9589
0.0199	9.0	126	0.2162	0.9464	0.9589
0.0171	10.0	140	0.3126	0.9286	0.9474
0.0121	11.0	154	0.2454	0.9286	0.9459
0.0106	12.0	168	0.2090	0.9464	0.9600
0.0115	13.0	182	0.1994	0.9464	0.9600
0.0084	14.0	196	0.2142	0.9464	0.9600
0.0065	15.0	210	0.2245	0.9464	0.9600
0.0066	16.0	224	0.2728	0.9286	0.9474
0.0027	17.0	238	0.2297	0.9464	0.9600
0.002	18.0	252	0.2121	0.9464	0.9600
0.0053	19.0	266	0.2454	0.9464	0.9600
0.0017	20.0	280	0.2457	0.9464	0.9600

Framework versions

• Transformers 4.53.3
• Pytorch 2.9.1+cu128
• Datasets 4.4.1
• Tokenizers 0.21.2