Edit model card

Whisper Tiny Belarusian

Repo to test model training

This model is a fine-tuned version of openai/whisper-tiny on the mozilla-foundation/common_voice_11_0 be dataset. It achieves the following results on the evaluation set:

  • Loss: 0.4388
  • Wer: 46.5201

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: 0.0001
  • 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: 10
  • training_steps: 300
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
2.5366 0.05 10 1.5402 94.5055
1.3721 0.1 20 1.0021 75.8242
0.9921 0.15 30 0.8322 75.0916
0.9844 0.2 40 0.8080 72.8938
0.7071 0.25 50 0.7862 77.2894
0.7998 0.3 60 0.7052 68.8645
0.6935 0.35 70 0.6781 64.2857
0.81 0.4 80 0.6341 63.5531
0.6133 0.45 90 0.6083 62.6374
0.6675 0.5 100 0.5851 62.8205
0.5577 0.55 110 0.5651 59.3407
0.6473 0.6 120 0.5638 58.0586
0.6018 0.65 130 0.5434 53.8462
0.5918 0.7 140 0.5385 54.9451
0.5654 0.75 150 0.5200 58.0586
0.587 0.8 160 0.4974 57.1429
0.6157 0.85 170 0.4834 53.2967
0.6803 0.9 180 0.4852 55.8608
0.4813 0.95 190 0.4686 51.2821
0.4952 1.0 200 0.4624 51.4652
0.3956 0.03 210 0.4690 52.0147
0.3719 0.07 220 0.4673 52.7473
0.3168 0.1 230 0.4499 51.4652
0.3582 0.13 240 0.4525 46.8864
0.2475 0.17 250 0.4612 52.3810
0.2988 0.2 260 0.4346 49.8168
0.2749 0.23 270 0.4249 48.9011
0.3368 0.27 280 0.4388 46.5201
0.2574 0.3 290 0.4309 46.7033
0.2921 0.33 300 0.4282 46.7033

Framework versions

  • Transformers 4.26.0.dev0
  • Pytorch 1.13.0+cu116
  • Datasets 2.7.1.dev0
  • Tokenizers 0.13.2
Downloads last month
50
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.

Dataset used to train ales/whisper-tiny-be-test

Evaluation results