Edit model card

whisper-small-mn-12

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2949
  • Wer: 32.3301
  • Cer: 13.3493

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: 32
  • eval_batch_size: 32
  • 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: 25000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer Cer
0.3012 1.05 1000 0.3749 43.2379 17.6739
0.2171 2.11 2000 0.3012 36.7435 15.2029
0.1732 3.16 3000 0.2823 33.4225 13.7561
0.145 4.21 4000 0.2822 32.4995 13.2436
0.1159 5.27 5000 0.2949 32.3301 13.3493
0.0863 6.32 6000 0.3116 32.7234 13.3892
0.0685 7.38 7000 0.3343 32.4776 13.3077
0.0506 8.43 8000 0.3584 33.3952 13.7736
0.0336 9.48 9000 0.3861 33.7011 13.8493
0.0215 10.54 10000 0.4193 33.7011 14.0140
0.0141 11.59 11000 0.4463 34.0343 14.0298
0.0089 12.64 12000 0.4660 33.6137 13.8052
0.0057 13.7 13000 0.4913 33.9797 13.9849
0.0039 14.75 14000 0.5078 33.9906 14.0656
0.0033 15.81 15000 0.5244 33.7721 13.9192
0.0024 16.86 16000 0.5358 33.7612 13.7910
0.0018 17.91 17000 0.5469 33.6465 13.8468
0.0013 18.97 18000 0.5614 33.6683 13.7553
0.0014 20.02 19000 0.5707 33.6574 13.8884
0.0006 21.07 20000 0.5835 34.0671 14.0764
0.0007 22.13 21000 0.5927 33.9742 14.0772
0.0005 23.18 22000 0.5994 34.0398 14.0290
0.0004 24.24 23000 0.6067 33.9469 13.9217
0.0003 25.29 24000 0.6109 33.9688 13.9591
0.0003 26.34 25000 0.6130 33.8267 13.8360

Framework versions

  • Transformers 4.26.0.dev0
  • Pytorch 1.13.0+cu117
  • Datasets 2.7.1.dev0
  • Tokenizers 0.13.2
Downloads last month
17
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Datasets used to train bayartsogt/whisper-small-mn-12

Evaluation results