Edit model card

b28-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: 0.3534
  • Wer: 0.2762

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: 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: 500
  • num_epochs: 30
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
5.6021 3.05 400 2.9592 1.0
1.8751 6.11 800 0.5165 0.5247
0.295 9.16 1200 0.3708 0.3901
0.1398 12.21 1600 0.3738 0.3419
0.0968 15.27 2000 0.3595 0.3202
0.0693 18.32 2400 0.3392 0.3048
0.0593 21.37 2800 0.3660 0.2951
0.0464 24.43 3200 0.3596 0.2809
0.0404 27.48 3600 0.3534 0.2762

Framework versions

  • Transformers 4.26.0
  • Pytorch 2.0.1+cu118
  • Datasets 2.14.4
  • Tokenizers 0.13.3
Downloads last month
0

Evaluation results