Edit model card

Whisper Medium Basque

This model is a fine-tuned version of openai/whisper-medium on the mozilla-foundation/common_voice_16_0 eu dataset. It achieves the following results on the evaluation set:

  • Loss: 0.1503
  • Wer: 9.1886

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: 4
  • 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: 8000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
0.4647 0.06 500 0.4529 34.2140
0.3163 0.12 1000 0.3516 26.0232
0.3232 0.19 1500 0.2996 21.1825
0.266 0.25 2000 0.2686 18.5126
0.2383 0.31 2500 0.2489 16.9412
0.1916 0.38 3000 0.2233 15.2831
0.2009 0.44 3500 0.2134 14.1419
0.2014 0.5 4000 0.2015 13.6579
0.1964 0.56 4500 0.1853 12.0198
0.1758 0.62 5000 0.1796 11.4651
0.2067 0.69 5500 0.1679 10.7989
0.213 0.75 6000 0.1618 10.3139
0.1272 1.03 6500 0.1551 9.8687
0.0744 1.09 7000 0.1534 9.5172
0.0726 1.16 7500 0.1518 9.3240
0.0627 1.22 8000 0.1503 9.1886

Framework versions

  • Transformers 4.26.0.dev0
  • Pytorch 1.13.1+cu117
  • Datasets 2.8.1.dev0
  • Tokenizers 0.13.2
Downloads last month
2
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 xezpeleta/whisper-medium-eu

Evaluation results