ns_finetune_urdu_asr_org

This model is a fine-tuned version of openai/whisper-large-v3 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1715
• Wer: 11.9535

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: 3e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 2500
• training_steps: 25972

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1999	0.0770	2000	0.2515	21.0816
0.2057	0.1540	4000	0.2292	18.5375
0.1721	0.2310	6000	0.2063	19.7264
0.1357	1.0580	8000	0.1906	15.0601
0.0982	1.1350	10000	0.1905	15.9422
0.0858	1.2120	12000	0.1808	16.5729
0.0777	2.0390	14000	0.1673	14.0714
0.0543	2.1160	16000	0.1777	13.1722
0.0426	2.1931	18000	0.1712	12.3626
0.0417	3.0201	20000	0.1682	12.5458
0.0286	3.0971	22000	0.1714	12.7418
0.0229	3.1741	24000	0.1715	11.9535

Framework versions

• Transformers 4.50.2
• Pytorch 2.6.0+cu124
• Datasets 3.4.1
• Tokenizers 0.21.1