Modelpublic

bagasshw/whisper-large-v2-javanese-openslr-v2

sourceHugging Faceapache-2.0updated 1y agoView on Hugging Face

0likes29downloads

Model Card

Whisper-Large-v2-Java-v2

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

—Loss: 0.1112
—Wer: 0.0612

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: 2e-05
—trainbatchsize: 2
—evalbatchsize: 2
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 4
—optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_ratio: 0.1
—training_steps: 20000
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4821	0.4	1000	0.3570	0.2268
0.4761	0.8	2000	0.3474	0.2397
0.3099	1.2	3000	0.3075	0.2026
0.3108	1.6	4000	0.2666	0.1971
0.3131	2.0	5000	0.2173	0.1531
0.1796	2.4	6000	0.2026	0.1384
0.17	2.8	7000	0.1922	0.1378
0.0995	3.2	8000	0.1792	0.1206
0.0972	3.6	9000	0.1670	0.1149
0.097	4.0	10000	0.1545	0.1096
0.0553	4.4	11000	0.1575	0.1030
0.0526	4.8	12000	0.1431	0.0890
0.0299	5.2	13000	0.1372	0.0833
0.0311	5.6	14000	0.1258	0.0780
0.0295	6.0	15000	0.1201	0.0725
0.0151	6.4	16000	0.1229	0.0742
0.0163	6.8	17000	0.1137	0.0658
0.0082	7.2	18000	0.1142	0.0639
0.0092	7.6	19000	0.1121	0.0627
0.006	8.0	20000	0.1112	0.0612

Framework versions

—Transformers 4.50.0.dev0
—Pytorch 2.7.0+cu128
—Datasets 2.16.0
—Tokenizers 0.21.1