README.md

metadata

license: mit
tags:
  - generated_from_trainer
metrics:
  - accuracy
  - precision
  - recall
  - f1
model-index:
  - name: indic-bert-finetuned-combined-DS
    results: []

indic-bert-finetuned-combined-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: 0.9783
• Accuracy: 0.5871
• Precision: 0.5527
• Recall: 0.5574
• F1: 0.5537

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-06
• 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: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
1.0904	1.0	711	1.0759	0.4452	0.4368	0.4155	0.3333
1.0537	2.0	1422	1.0363	0.5063	0.5079	0.4902	0.4889
1.0233	3.0	2133	1.0207	0.5302	0.5195	0.5266	0.4956
1.0091	3.99	2844	1.0168	0.5379	0.5248	0.5313	0.5124
0.9983	4.99	3555	1.0009	0.5681	0.5344	0.5424	0.5335
0.9854	5.99	4266	0.9950	0.5829	0.5490	0.5548	0.5492
0.9728	6.99	4977	0.9917	0.5751	0.5436	0.5515	0.5434
0.9616	7.99	5688	0.9888	0.5492	0.5183	0.5308	0.5107
0.9476	8.99	6399	0.9815	0.5836	0.5488	0.5526	0.5499
0.9355	9.99	7110	0.9962	0.5520	0.5316	0.5419	0.5223
0.924	10.98	7821	0.9823	0.5674	0.5363	0.5453	0.5315
0.9112	11.98	8532	0.9773	0.5829	0.5479	0.5504	0.5488
0.9002	12.98	9243	0.9761	0.5815	0.5452	0.5479	0.5459
0.8904	13.98	9954	0.9726	0.5913	0.5558	0.5495	0.5512
0.8823	14.98	10665	0.9785	0.5843	0.5526	0.5583	0.5533
0.8742	15.98	11376	0.9765	0.5843	0.5493	0.5543	0.5500
0.8653	16.98	12087	0.9746	0.5864	0.5501	0.5532	0.5508
0.8612	17.97	12798	0.9770	0.5885	0.5539	0.5566	0.5548
0.8558	18.97	13509	0.9796	0.5836	0.5510	0.5565	0.5520
0.8561	19.97	14220	0.9783	0.5871	0.5527	0.5574	0.5537

Framework versions

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