Whisper Base Urdu - Ali Abbas

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

  • Loss: 0.2081
  • Wer: 34.3796

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: 16
  • eval_batch_size: 8
  • 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: 4000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
0.4518 0.3521 100 0.4355 63.5879
0.3641 0.7042 200 0.3525 55.6127
0.3096 1.0563 300 0.3102 50.6367
0.2726 1.4085 400 0.2806 47.1325
0.2466 1.7606 500 0.2573 44.3656
0.2203 2.1127 600 0.2406 41.8900
0.1953 2.4648 700 0.2288 40.7312
0.1953 2.8169 800 0.2183 38.2804
0.1561 3.1690 900 0.2124 37.6545
0.1657 3.5211 1000 0.2066 36.4864
0.1597 3.8732 1100 0.2016 35.3277
0.1282 4.2254 1200 0.1998 35.3741
0.1363 4.5775 1300 0.1963 35.4919
0.1299 4.9296 1400 0.1941 34.4260
0.1203 5.2817 1500 0.1946 33.9892
0.1147 5.6338 1600 0.1929 34.2308
0.1073 5.9859 1700 0.1909 33.5957
0.0968 6.3380 1800 0.1926 33.7847
0.0941 6.6901 1900 0.1930 33.6050
0.0899 7.0423 2000 0.1918 33.8218
0.0834 7.3944 2100 0.1934 33.3199
0.0796 7.7465 2200 0.1937 33.9396
0.0673 8.0986 2300 0.1947 33.1960
0.078 8.4507 2400 0.1962 33.9055
0.0695 8.8028 2500 0.1952 34.1689
0.0649 9.1549 2600 0.1976 33.0906
0.0611 9.5070 2700 0.1981 33.9024
0.0616 9.8592 2800 0.1986 33.6235
0.0584 10.2113 2900 0.2010 34.0139
0.0548 10.5634 3000 0.2017 33.8095
0.0558 10.9155 3100 0.2020 34.3641
0.0489 11.2676 3200 0.2033 34.2525
0.0488 11.6197 3300 0.2045 34.3888
0.0491 11.9718 3400 0.2048 34.1751
0.0465 12.3239 3500 0.2066 34.5252
0.0431 12.6761 3600 0.2067 34.1441
0.0444 13.0282 3700 0.2072 34.1255
0.0425 13.3803 3800 0.2079 34.4415
0.0446 13.7324 3900 0.2081 34.1689
0.0414 14.0845 4000 0.2081 34.3796

Framework versions

  • Transformers 4.50.0.dev0
  • Pytorch 2.6.0+cu124
  • Datasets 3.4.0
  • Tokenizers 0.21.1
Downloads last month
4
Safetensors
Model size
72.6M params
Tensor type
F32
·
Inference Providers NEW
This model is not currently available via any of the supported Inference Providers.

Model tree for m-aliabbas1/whisper-base-urdu

Finetuned
(438)
this model

Evaluation results