Edit model card

whisper-tiny-finetune

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

  • Loss: 0.5792
  • Wer: 20.6820

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: 128
  • 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: 500
  • training_steps: 1000

Training results

Training Loss Epoch Step Validation Loss Wer
4.1356 0.2778 10 4.1201 47.9002
4.0312 0.5556 20 4.0231 47.3319
3.917 0.8333 30 3.8659 46.5425
3.7606 1.1111 40 3.6569 45.8478
3.4823 1.3889 50 3.3969 44.5216
3.0938 1.6667 60 3.0765 41.9324
2.7895 1.9444 70 2.6692 34.9542
2.3101 2.2222 80 2.1389 34.5122
1.6935 2.5 90 1.5546 34.8911
1.1419 2.7778 100 1.0650 36.5330
0.904 3.0556 110 0.8400 29.4601
0.7536 3.3333 120 0.7657 28.9233
0.6857 3.6111 130 0.7202 27.7550
0.6609 3.8889 140 0.6886 26.6814
0.5804 4.1667 150 0.6656 25.6710
0.5611 4.4444 160 0.6465 25.0710
0.5574 4.7222 170 0.6293 24.3448
0.552 5.0 180 0.6135 24.0606
0.4717 5.2778 190 0.6024 24.5974
0.4681 5.5556 200 0.5898 24.0290
0.4679 5.8333 210 0.5778 23.5238
0.4351 6.1111 220 0.5670 23.6501
0.3982 6.3889 230 0.5599 23.2081
0.3892 6.6667 240 0.5520 22.0714
0.3771 6.9444 250 0.5439 21.1872
0.3532 7.2222 260 0.5372 21.6925
0.3435 7.5 270 0.5309 27.5024
0.336 7.7778 280 0.5253 20.9346
0.3088 8.0556 290 0.5201 20.4610
0.3014 8.3333 300 0.5184 20.5242
0.316 8.6111 310 0.5146 20.2400
0.2931 8.8889 320 0.5118 19.9874
0.2228 9.1667 330 0.5079 20.3663
0.2445 9.4444 340 0.5052 20.2716
0.2343 9.7222 350 0.5039 20.2084
0.2893 10.0 360 0.5023 20.0189
0.2014 10.2778 370 0.5030 20.0505
0.2048 10.5556 380 0.5036 19.6400
0.1941 10.8333 390 0.5003 20.1137
0.1601 11.1111 400 0.4992 19.8295
0.1647 11.3889 410 0.5010 19.8926
0.1519 11.6667 420 0.5044 19.6716
0.1747 11.9444 430 0.5005 20.1137
0.1194 12.2222 440 0.5076 20.7452
0.1021 12.5 450 0.5104 19.9242
0.1115 12.7778 460 0.5102 20.7136
0.1355 13.0556 470 0.5068 20.3979
0.0824 13.3333 480 0.5152 20.5557
0.0858 13.6111 490 0.5189 20.3663
0.0786 13.8889 500 0.5225 21.1557
0.0564 14.1667 510 0.5250 20.9031
0.056 14.4444 520 0.5232 20.8715
0.0558 14.7222 530 0.5282 20.5557
0.0657 15.0 540 0.5299 20.7452
0.0369 15.2778 550 0.5342 20.6505
0.0355 15.5556 560 0.5341 20.1137
0.0383 15.8333 570 0.5370 20.4926
0.0333 16.1111 580 0.5401 20.5557
0.027 16.3889 590 0.5455 20.9346
0.0261 16.6667 600 0.5480 20.6189
0.024 16.9444 610 0.5494 20.4294
0.0164 17.2222 620 0.5505 20.3663
0.0159 17.5 630 0.5577 20.7136
0.0168 17.7778 640 0.5549 20.9031
0.015 18.0556 650 0.5555 20.8083
0.0116 18.3333 660 0.5596 20.9978
0.0131 18.6111 670 0.5614 20.9346
0.0121 18.8889 680 0.5634 20.3663
0.009 19.1667 690 0.5643 20.7452
0.0108 19.4444 700 0.5633 20.3031
0.0096 19.7222 710 0.5666 20.3979
0.0123 20.0 720 0.5660 20.4610
0.009 20.2778 730 0.5695 20.5242
0.0099 20.5556 740 0.5684 20.3663
0.0079 20.8333 750 0.5701 20.7768
0.008 21.1111 760 0.5701 20.7136
0.0084 21.3889 770 0.5719 20.7136
0.0076 21.6667 780 0.5724 20.4610
0.0081 21.9444 790 0.5724 20.7136
0.0067 22.2222 800 0.5731 20.6820
0.0076 22.5 810 0.5737 20.4926
0.0079 22.7778 820 0.5748 20.3979
0.0069 23.0556 830 0.5747 20.6820
0.0066 23.3333 840 0.5751 20.7136
0.0062 23.6111 850 0.5755 20.7136
0.0071 23.8889 860 0.5764 20.5873
0.0062 24.1667 870 0.5774 20.7136
0.0059 24.4444 880 0.5769 20.5873
0.0066 24.7222 890 0.5772 20.6189
0.0066 25.0 900 0.5778 20.5873
0.0066 25.2778 910 0.5779 20.5557
0.0062 25.5556 920 0.5781 20.5873
0.006 25.8333 930 0.5787 20.6189
0.0061 26.1111 940 0.5789 20.5873
0.0056 26.3889 950 0.5788 20.5557
0.006 26.6667 960 0.5789 20.5873
0.0055 26.9444 970 0.5790 20.5873
0.0057 27.2222 980 0.5791 20.6189
0.0063 27.5 990 0.5792 20.6820
0.0059 27.7778 1000 0.5792 20.6820

Framework versions

  • Transformers 4.40.1
  • Pytorch 2.2.1+cu121
  • Datasets 2.19.1.dev0
  • Tokenizers 0.19.1
Downloads last month
2
Safetensors
Model size
37.8M params
Tensor type
F32
·

Finetuned from