Modelpublic

Roshana/CV30_finetuning

sourceHugging Faceupdated 4y agoView on Hugging Face

0likes8downloads

Model Card

CV30_finetuning

This model is a fine-tuned version of Roshana/CV11_finetuning1 on the None dataset. It achieves the following results on the evaluation set:

—Loss: 0.7956
—Wer: 0.3677
—Cer: 0.1436

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: 0.0003
—trainbatchsize: 8
—evalbatchsize: 8
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 16
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 500
—num_epochs: 10
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.1203	0.86	400	0.7747	0.4192	0.1656
0.2191	1.72	800	0.7546	0.4356	0.1771
0.213	2.59	1200	0.6965	0.4276	0.1709
0.1858	3.45	1600	0.7189	0.4156	0.1673
0.1597	4.31	2000	0.7462	0.4031	0.1606
0.1398	5.17	2400	0.7622	0.3944	0.1584
0.1245	6.03	2800	0.7527	0.3917	0.1553
0.1035	6.9	3200	0.7347	0.3819	0.1532
0.0918	7.76	3600	0.7825	0.3786	0.1497
0.0767	8.62	4000	0.7909	0.3724	0.1444
0.0722	9.48	4400	0.7956	0.3677	0.1436

Framework versions

—Transformers 4.25.1
—Pytorch 1.13.0+cu116
—Datasets 2.7.1
—Tokenizers 0.13.2