hing-roberta-ours-run-5

This model is a fine-tuned version of l3cube-pune/hing-roberta on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 3.0980
• Accuracy: 0.725
• Precision: 0.6881
• Recall: 0.6575
• F1: 0.6651

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
0.9336	1.0	200	0.7394	0.675	0.6450	0.6509	0.6398
0.6924	2.0	400	0.9530	0.66	0.6285	0.5845	0.5551
0.4406	3.0	600	0.8914	0.68	0.6462	0.6527	0.6479
0.2493	4.0	800	1.7083	0.68	0.6441	0.6446	0.6426
0.1231	5.0	1000	1.9496	0.695	0.6570	0.6624	0.6591
0.0788	6.0	1200	2.5025	0.67	0.6209	0.6039	0.6011
0.0408	7.0	1400	2.2651	0.695	0.6594	0.6617	0.6517
0.0434	8.0	1600	2.4072	0.725	0.6941	0.6754	0.6710
0.0074	9.0	1800	2.7817	0.7	0.6535	0.6467	0.6488
0.023	10.0	2000	2.8578	0.7	0.6470	0.6353	0.6337
0.0151	11.0	2200	2.7783	0.695	0.6457	0.6373	0.6390
0.0108	12.0	2400	2.5953	0.695	0.6563	0.6586	0.6564
0.0192	13.0	2600	3.0715	0.705	0.6631	0.6326	0.6320
0.0149	14.0	2800	3.1048	0.715	0.6769	0.6450	0.6503
0.0205	15.0	3000	2.7812	0.71	0.6657	0.6538	0.6565
0.0024	16.0	3200	2.9304	0.72	0.6796	0.6537	0.6610
0.0033	17.0	3400	2.7170	0.73	0.6899	0.6760	0.6811
0.0056	18.0	3600	2.9693	0.72	0.6783	0.6560	0.6628
0.0015	19.0	3800	3.0943	0.72	0.6825	0.6541	0.6611
0.0017	20.0	4000	3.0980	0.725	0.6881	0.6575	0.6651

Framework versions

• Transformers 4.25.1
• Pytorch 1.13.0+cu116
• Tokenizers 0.13.2