sutra-classifier-adv-epoch50-cosine

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

• Loss: 3.6188
• Accuracy: 0.8627
• F1: 0.8625
• Adv Acc: 0.0157

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: 8.8e-05
• train_batch_size: 32
• eval_batch_size: 64
• 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: cosine
• lr_scheduler_warmup_ratio: 0.079905
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Adv Acc
0.783	1.0	72	0.7087	0.8549	0.8549	0.1451
0.7764	2.0	144	3.0179	0.8863	0.8862	0.1451
2.9074	3.0	216	2.7534	0.8941	0.8941	0.1451
3.3278	4.0	288	4.3535	0.8549	0.8546	0.0
3.6887	5.0	360	6.0152	0.9059	0.9055	0.1216
5.4634	6.0	432	6.3364	0.8863	0.8851	0.0118
3.965	7.0	504	3.9418	0.8392	0.8391	0.1294
6.9282	8.0	576	3.6172	0.8667	0.8661	0.0706
3.7356	9.0	648	3.3176	0.8706	0.8694	0.1059
2.9881	10.0	720	2.9793	0.8745	0.8739	0.1608
2.9243	11.0	792	3.2793	0.8314	0.8305	0.1098
2.8977	12.0	864	3.6846	0.8863	0.8862	0.1529
3.2782	13.0	936	4.2865	0.8824	0.8822	0.0
3.2658	14.0	1008	3.6497	0.8627	0.8627	0.0
3.121	15.0	1080	3.6188	0.8627	0.8625	0.0157

Framework versions

• Transformers 4.49.0
• Pytorch 2.7.1+cu126
• Datasets 4.0.0
• Tokenizers 0.21.4

Wandb training stats

https://wandb.ai/bhaswata08/huggingface/runs/jq7gvfq7