indic-bert-finetuned-ours-DS

This model is a fine-tuned version of ai4bharat/indic-bert on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 2.1832
• Accuracy: 0.655
• Precision: 0.6023
• Recall: 0.6027
• F1: 0.6025

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 32
• seed: 43
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 25

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
1.0681	0.99	99	1.0180	0.365	0.3435	0.4038	0.2773
0.9384	1.98	198	0.8475	0.62	0.6235	0.5610	0.4821
0.8201	2.97	297	0.8187	0.68	0.6839	0.6086	0.5812
0.7178	3.96	396	0.7717	0.7	0.7117	0.6670	0.6470
0.62	4.95	495	0.7839	0.66	0.6165	0.6244	0.6174
0.5135	5.94	594	0.8392	0.675	0.6270	0.6234	0.6246
0.4073	6.93	693	0.8930	0.665	0.6251	0.6254	0.6240
0.3365	7.92	792	0.9362	0.675	0.6298	0.6276	0.6242
0.2719	8.91	891	1.0108	0.685	0.6388	0.6293	0.6326
0.2007	9.9	990	1.1214	0.675	0.6300	0.6299	0.6290
0.1567	10.89	1089	1.1367	0.67	0.6193	0.6212	0.6178
0.1074	11.88	1188	1.3157	0.65	0.6292	0.6317	0.6227
0.0821	12.87	1287	1.5412	0.665	0.6415	0.6330	0.6259
0.0588	13.86	1386	1.7215	0.64	0.5862	0.5869	0.5865
0.0337	14.85	1485	1.7556	0.64	0.6078	0.6082	0.6032
0.0244	15.84	1584	1.8713	0.66	0.6173	0.6186	0.6158
0.0166	16.83	1683	1.9666	0.66	0.5995	0.5973	0.5973
0.0124	17.82	1782	1.9245	0.66	0.6165	0.6194	0.6163
0.0079	18.81	1881	2.0814	0.65	0.6026	0.6023	0.6012
0.0051	19.8	1980	2.1029	0.645	0.6014	0.5986	0.5975
0.0031	20.79	2079	2.1155	0.655	0.6029	0.6027	0.6023
0.0029	21.78	2178	2.1221	0.655	0.6	0.6000	0.5999
0.0021	22.77	2277	2.2065	0.65	0.5917	0.5898	0.5905
0.0017	23.76	2376	2.1903	0.65	0.5910	0.5898	0.5902
0.0016	24.75	2475	2.1832	0.655	0.6023	0.6027	0.6025

Framework versions

• Transformers 4.20.1
• Pytorch 1.10.1+cu111
• Datasets 2.3.2
• Tokenizers 0.12.1