Modelpublic

kaarthu2003/whisper-small-telugu_learning-rate-1e-5-batch-size-32

sourceHugging Faceapache-2.0updated 1y agoView on Hugging Face

0likes2downloads

Model Card

whisper-small-telugu_learning-rate-1e-5-batch-size-32

This model is a fine-tuned version of openai/whisper-small on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 0.2913
—Wer: 0.3679
—Cer: 0.0992

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
—trainbatchsize: 16
—evalbatchsize: 16
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 32
—optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_steps: 400
—num_epochs: 20
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.5388	1.0900	400	0.1649	0.5455	0.1796
0.1218	2.1801	800	0.1323	0.5317	0.1889
0.0883	3.2701	1200	0.1302	0.3957	0.1174
0.0675	4.3602	1600	0.1365	0.4234	0.1324
0.0505	5.4502	2000	0.1449	0.3825	0.1042
0.0389	6.5402	2400	0.1614	0.3788	0.1032
0.0267	7.6303	2800	0.1790	0.3732	0.1007
0.0182	8.7203	3200	0.1951	0.3682	0.0990
0.0111	9.8104	3600	0.2162	0.3672	0.0976
0.0068	10.9004	4000	0.2250	0.3659	0.0973
0.004	11.9905	4400	0.2386	0.3731	0.1005
0.0021	13.0791	4800	0.2513	0.3663	0.0980
0.0011	14.1692	5200	0.2658	0.3646	0.0966
0.0006	15.2592	5600	0.2718	0.3682	0.0973
0.0004	16.3492	6000	0.2851	0.3693	0.0986
0.0003	17.4393	6400	0.2854	0.3667	0.0983
0.0002	18.5293	6800	0.2886	0.3670	0.0991
0.0001	19.6194	7200	0.2913	0.3679	0.0992

Framework versions

—Transformers 4.51.2
—Pytorch 2.6.0+cu124
—Datasets 3.5.0
—Tokenizers 0.21.1