Edit model card

whisper-small-mn-11

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: 1.0138
  • Wer: 51.9664
  • Cer: 19.8770

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: 15000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer Cer
0.0025 32.26 1000 0.8913 53.6214 20.6866
0.0007 64.52 2000 0.9326 52.1685 19.5991
0.0001 96.77 3000 1.0138 51.9664 19.8770
0.0001 129.03 4000 1.0639 52.1248 19.9178
0.0 161.29 5000 1.1236 52.0428 19.9652
0.0 193.55 6000 1.1677 52.4634 20.0351
0.0 225.81 7000 1.2224 52.5836 20.1258
0.0 258.06 8000 1.2633 52.7310 20.2073
0.0 290.32 9000 1.3152 52.8184 20.2273
0.0 322.58 10000 1.3530 52.9495 20.3080
0.0 354.84 11000 1.3995 53.0260 20.3088
0.0 387.1 12000 1.4306 52.9878 20.2057
0.0 419.35 13000 1.4674 52.9714 20.3113
0.0 451.61 14000 1.4859 52.9386 20.2947
0.0 483.87 15000 1.4994 52.9768 20.3280

Framework versions

  • Transformers 4.26.0.dev0
  • Pytorch 1.13.0+cu117
  • Datasets 2.7.1.dev0
  • Tokenizers 0.13.2
Downloads last month
2

Datasets used to train bayartsogt/whisper-small-mn-11

Evaluation results