meg51/whisper-medium-hindi-15000-1

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: 0.3821
• Wer: 18.5865

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: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0669	3.3898	1000	0.2078	20.8512
0.0131	6.7797	2000	0.2584	20.0312
0.002	10.1695	3000	0.3048	19.2698
0.0024	13.5593	4000	0.3192	19.1429
0.0025	16.9492	5000	0.3127	19.0941
0.0008	20.3390	6000	0.3412	19.1429
0.0008	23.7288	7000	0.3438	18.3913
0.0011	27.1186	8000	0.3465	18.8501
0.001	30.5085	9000	0.3549	18.5377
0.0002	33.8983	10000	0.3551	18.0594
0.0	37.2881	11000	0.3689	18.3522
0.0	40.6780	12000	0.3721	18.3229
0.0	44.0678	13000	0.3821	18.5865

Framework versions

• Transformers 4.43.3
• Pytorch 2.4.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1