Modelpublic

tranha/whisper-finetuned-v3_2k5step

sourceHugging Facemitupdated 1y agoView on Hugging Face

0likes3downloads

Model Card

whisper-finetuned-v3_2k5step

This model is a fine-tuned version of openai/whisper-large-v3-turbo on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 0.0569
—Wer: 51.4499

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: 2
—evalbatchsize: 2
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 4
—optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_steps: 500
—training_steps: 2500
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.6705	0.1118	100	1.2111	158.0404
1.0129	0.2236	200	0.6581	464.8067
0.415	0.3354	300	0.2700	101.1863
0.247	0.4472	400	0.2067	76.8014
0.2215	0.5590	500	0.1855	86.9069
0.1739	0.6708	600	0.1678	73.7698
0.1528	0.7826	700	0.1437	70.7821
0.1452	0.8944	800	0.1452	74.2091
0.1241	1.0056	900	0.1157	65.7293
0.0991	1.1174	1000	0.1029	66.3005
0.0837	1.2292	1100	0.1048	61.9069
0.0824	1.3410	1200	0.0945	61.4236
0.081	1.4528	1300	0.0913	62.3023
0.0725	1.5646	1400	0.0830	58.5237
0.0715	1.6764	1500	0.0774	60.8524
0.0704	1.7881	1600	0.0732	58.6116
0.0631	1.8999	1700	0.0723	57.3814
0.064	2.0112	1800	0.0677	56.6784
0.0394	2.1230	1900	0.0697	54.8770
0.0378	2.2348	2000	0.0656	55.4482
0.0292	2.3466	2100	0.0625	52.7241
0.0355	2.4584	2200	0.0599	54.7452
0.032	2.5702	2300	0.0583	53.0756
0.0299	2.6819	2400	0.0574	51.4060
0.0256	2.7937	2500	0.0569	51.4499

Framework versions

—Transformers 4.51.3
—Pytorch 2.7.0+cu126
—Datasets 3.5.1
—Tokenizers 0.21.1