meg51/whisper-medium-hindi-20000

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: 0.4333
• Wer: 18.0203

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: 25000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0669	3.3898	1000	0.2086	20.9684
0.0115	6.7797	2000	0.2637	19.7579
0.0034	10.1695	3000	0.3012	19.6408
0.0026	13.5593	4000	0.3179	19.2893
0.0014	16.9492	5000	0.3242	18.7817
0.0024	20.3390	6000	0.3348	19.1624
0.0024	23.7288	7000	0.3421	19.7774
0.0006	27.1186	8000	0.3511	18.6939
0.0008	30.5085	9000	0.3632	18.8989
0.0007	33.8983	10000	0.3600	18.7622
0.0006	37.2881	11000	0.3470	18.4791
0.0002	40.6780	12000	0.3548	18.2936
0.0001	44.0678	13000	0.3711	18.0594
0.0006	47.4576	14000	0.3733	18.2839
0.0003	50.8475	15000	0.3766	18.1667
0.0	54.2373	16000	0.3745	18.0203
0.0	57.6271	17000	0.3914	17.8739
0.0	61.0169	18000	0.4003	17.9032
0.0	64.4068	19000	0.4081	17.8641
0.0	67.7966	20000	0.4153	17.8544
0.0	71.1864	21000	0.4219	17.8544
0.0	74.5763	22000	0.4281	18.0105
0.0	77.9661	23000	0.4333	18.0203

Framework versions

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