Edit model card

wav2vec2-common_voice-tr-demo

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the COMMON_VOICE - TR dataset. It achieves the following results on the evaluation set:

  • Loss: 0.3794
  • Wer: 0.3446

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: 0.0003
  • train_batch_size: 16
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • num_epochs: 15.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
No log 0.92 100 3.5956 1.0
No log 1.83 200 3.0269 0.9999
No log 2.75 300 0.9827 0.8111
No log 3.67 400 0.6236 0.6304
3.1866 4.59 500 0.5016 0.5264
3.1866 5.5 600 0.4523 0.4935
3.1866 6.42 700 0.4306 0.4528
3.1866 7.34 800 0.4328 0.4329
3.1866 8.26 900 0.4026 0.4105
0.227 9.17 1000 0.4096 0.4080
0.227 10.09 1100 0.3921 0.3915
0.227 11.01 1200 0.3830 0.3778
0.227 11.93 1300 0.3846 0.3616
0.227 12.84 1400 0.3888 0.3619
0.1046 13.76 1500 0.3861 0.3509
0.1046 14.68 1600 0.3798 0.3455

Framework versions

  • Transformers 4.26.0.dev0
  • Pytorch 1.12.0+cu116
  • Datasets 2.7.1
  • Tokenizers 0.13.2
Downloads last month
2

Dataset used to train Jaewan/wav2vec2-common_voice-tr-demo

Evaluation results