meg51/whisper-medium-hindi-15000-2

Whisper Medium Hindi -megha sharma

This model is a fine-tuned version of openai/whisper-medium on the Google Fleurs dataset. It achieves the following results on the evaluation set:

• Loss: 0.4008
• Wer: 17.7470

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: 5e-06
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• training_steps: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.067	3.3898	1000	0.2071	20.8024
0.0116	6.7797	2000	0.2594	19.6505
0.0032	10.1695	3000	0.2891	19.0062
0.0029	13.5593	4000	0.3075	18.9086
0.0026	16.9492	5000	0.3211	19.1722
0.0033	20.3390	6000	0.3254	18.6841
0.0014	23.7288	7000	0.3304	18.2546
0.0008	27.1186	8000	0.3422	18.4889
0.0023	30.5085	9000	0.3379	18.0886
0.0009	33.8983	10000	0.3525	18.4010
0.0006	37.2881	11000	0.3511	18.0301
0.0001	40.6780	12000	0.3651	18.1863
0.0001	44.0678	13000	0.3627	17.8446
0.0	47.4576	14000	0.3775	17.6982
0.0	50.8475	15000	0.3868	17.7079
0.0	54.2373	16000	0.3944	17.7079
0.0	57.6271	17000	0.4008	17.7470

Framework versions

• Transformers 4.44.0
• Pytorch 2.4.0+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1