whisper-small-hindi-transcribe

This model is a fine-tuned version of openai/whisper-small on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3134
• Wer: 21.0323

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-05
• train_batch_size: 64
• eval_batch_size: 16
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 8100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1415	1.8473	750	0.1538	22.5756
0.0787	3.6946	1500	0.1400	20.1356
0.039	5.5419	2250	0.1561	20.1668
0.0163	7.3892	3000	0.1859	20.6361
0.0057	9.2365	3750	0.2193	21.0532
0.0018	11.0837	4500	0.2519	20.9906
0.0011	12.9310	5250	0.2728	21.1470
0.0006	14.7783	6000	0.2896	21.3139
0.0002	16.6256	6750	0.3043	21.0636
0.0002	18.4729	7500	0.3134	21.0323

Framework versions

• Transformers 4.48.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0