whisper-small-sagale_dataset-ormo-10hrs-v5

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

  • Wer: 0.3355
  • Cer: 0.1622
  • Loss: 0.1740

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-05
  • train_batch_size: 16
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Use adamw_bnb_8bit with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.05
  • num_epochs: 100
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Wer Cer Validation Loss
1.6671 1.0 196 0.9903 0.6375 0.8798
0.2728 2.0 392 0.4929 0.2124 0.1525
0.1106 3.0 588 0.4350 0.1909 0.1279
0.076 4.0 784 0.4198 0.1929 0.1284
0.062 5.0 980 0.4264 0.1937 0.1381
0.0525 6.0 1176 0.3942 0.1791 0.1289
0.0399 7.0 1372 0.4073 0.1897 0.1321
0.032 8.0 1568 0.3698 0.1702 0.1303
0.0246 9.0 1764 0.3831 0.1746 0.1446
0.0214 10.0 1960 0.3674 0.1705 0.1477
0.0164 11.0 2156 0.3711 0.1713 0.1485
0.0133 12.0 2352 0.3617 0.1707 0.1479
0.0128 13.0 2548 0.3609 0.1722 0.1496
0.0117 14.0 2744 0.3628 0.1739 0.1486
0.0089 15.0 2940 0.3526 0.1674 0.1521
0.0088 16.0 3136 0.3543 0.1687 0.1598
0.007 17.0 3332 0.3588 0.1684 0.1673
0.0069 18.0 3528 0.3556 0.1671 0.1722
0.0065 19.0 3724 0.3583 0.1672 0.1751
0.0062 20.0 3920 0.3564 0.1660 0.1612
0.0056 21.0 4116 0.3473 0.1649 0.1659
0.0043 22.0 4312 0.3545 0.1659 0.1755
0.0049 23.0 4508 0.3499 0.1664 0.1664
0.0049 24.0 4704 0.3615 0.1686 0.1666
0.0043 25.0 4900 0.3528 0.1656 0.1720
0.0036 26.0 5096 0.3418 0.1638 0.1650
0.0035 27.0 5292 0.3490 0.1634 0.1826
0.0037 28.0 5488 0.3509 0.1679 0.1696
0.0027 29.0 5684 0.3439 0.1633 0.1814
0.0028 30.0 5880 0.3416 0.1651 0.1776
0.0032 31.0 6076 0.3473 0.1650 0.1674
0.0028 32.0 6272 0.3414 0.1632 0.1771
0.0028 33.0 6468 0.3357 0.1624 0.1840
0.0026 34.0 6664 0.3416 0.1647 0.1787
0.0031 35.0 6860 0.3482 0.1650 0.1855
0.0025 36.0 7056 0.3454 0.1631 0.1821
0.0025 37.0 7252 0.3475 0.1653 0.1707
0.0022 38.0 7448 0.3374 0.1617 0.1713
0.0016 39.0 7644 0.3414 0.1627 0.1781
0.0023 40.0 7840 0.3444 0.1630 0.1767
0.0022 41.0 8036 0.3374 0.1638 0.1697
0.0014 42.0 8232 0.3397 0.1632 0.1907
0.0018 43.0 8428 0.3355 0.1622 0.1740

Framework versions

  • Transformers 4.49.0
  • Pytorch 2.5.1+cu121
  • Datasets 2.20.0
  • Tokenizers 0.21.0
Downloads last month
14
Safetensors
Model size
242M params
Tensor type
F32
·
Inference Providers NEW
This model is not currently available via any of the supported Inference Providers.

Model tree for KYAGABA/whisper-small-sagale_dataset-ormo-10hrs-v5

Finetuned
(2395)
this model