Edit model card

b21-wav2vec2-large-xls-r-romansh-colab

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:

  • Loss: 0.8091
  • Wer: 0.6304

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: 0.0004
  • train_batch_size: 4
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 8
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 100
  • num_epochs: 30
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
6.5829 0.76 100 2.9564 1.0
2.9568 1.52 200 3.0768 1.0
2.9578 2.29 300 3.0654 1.0
2.957 3.05 400 2.9377 1.0
2.9419 3.81 500 2.9408 1.0
2.9567 4.58 600 2.9395 1.0
2.9625 5.34 700 2.9388 1.0
2.9395 6.11 800 2.9374 1.0
2.9285 6.87 900 2.9240 1.0
2.9187 7.63 1000 2.9057 1.0
2.9251 8.4 1100 2.8985 1.0
2.9033 9.16 1200 2.8942 1.0
2.8877 9.92 1300 2.8917 1.0
2.8586 10.68 1400 2.7719 1.0
2.5777 11.45 1500 2.2424 1.0
1.9243 12.21 1600 1.7068 0.9772
1.4534 12.97 1700 1.2780 0.9585
1.1793 13.74 1800 1.1482 0.9360
1.0026 14.5 1900 1.0673 0.8852
0.8879 15.27 2000 0.9651 0.8433
0.7933 16.03 2100 0.8973 0.8216
0.6895 16.79 2200 0.8396 0.8034
0.6531 17.56 2300 0.8131 0.7713
0.5753 18.32 2400 0.8388 0.7531
0.5621 19.08 2500 0.7844 0.7632
0.5076 19.84 2600 0.7629 0.7485
0.4672 20.61 2700 0.7777 0.7497
0.443 21.37 2800 0.8001 0.7292
0.4129 22.14 2900 0.7902 0.7094
0.3767 22.9 3000 0.7569 0.6784
0.357 23.66 3100 0.7726 0.6903
0.3378 24.43 3200 0.8016 0.6882
0.3199 25.19 3300 0.7854 0.6677
0.3144 25.95 3400 0.7792 0.6509
0.3025 26.71 3500 0.8157 0.6695
0.2919 27.48 3600 0.8215 0.6633
0.2762 28.24 3700 0.8167 0.6500
0.2679 29.01 3800 0.8144 0.6311
0.2671 29.77 3900 0.8091 0.6304

Framework versions

  • Transformers 4.26.0
  • Pytorch 2.0.1+cu118
  • Datasets 2.14.4
  • Tokenizers 0.13.3
Downloads last month
1

Evaluation results