Modelpublic

miosipof/whisper_medium_BB_and_EC_v2.1

sourceHugging Faceupdated 2y agoView on Hugging Face

0likes7downloads

Model Card

Whisper Medium

This model is a fine-tuned version of b-brave/asr_double_training_15-10-2024_merged on the ASRBBand_EC dataset. It achieves the following results on the evaluation set:

—Loss: 0.4620
—Wer: 36.9269

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
—trainbatchsize: 16
—evalbatchsize: 16
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 32
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: reducelron_plateau
—lrschedulerwarmup_steps: 100
—num_epochs: 12
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7705	0.8929	100	0.4885	36.5551
0.7193	1.7857	200	0.4840	36.6791
0.7376	2.6786	300	0.4808	36.4312
0.6975	3.5714	400	0.4783	36.4312
0.6499	4.4643	500	0.4763	35.8116
0.7137	5.3571	600	0.4744	35.9356
0.6397	6.25	700	0.4727	35.9356
0.6441	7.1429	800	0.4708	35.9356
0.6756	8.0357	900	0.4690	35.9356
0.6331	8.9286	1000	0.4673	36.3073
0.6411	9.8214	1100	0.4656	36.3073
0.6029	10.7143	1200	0.4638	36.6791
0.6229	11.6071	1300	0.4620	36.9269

Framework versions

—PEFT 0.13.2
—Transformers 4.45.2
—Pytorch 2.2.0
—Datasets 3.1.0
—Tokenizers 0.20.3