salti's picture
add model
fb12feb
metadata
model-index:
  - name: wav2vec2-large-xlsr-arabic-common_voice-10_epochs

wav2vec2-large-xlsr-arabic-common_voice-10_epochs

This model was trained from scratch on an unkown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.3581
  • Wer: 0.4555

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_ratio: 0.1
  • num_epochs: 10
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
3.1701 0.9 400 3.1599 1.0
0.8933 1.8 800 0.7198 0.7877
0.5849 2.7 1200 0.5046 0.6253
0.3858 3.6 1600 0.4247 0.5561
0.3083 4.49 2000 0.4026 0.5251
0.2556 5.39 2400 0.4010 0.5051
0.2221 6.29 2800 0.3765 0.4861
0.2026 7.19 3200 0.3652 0.4794
0.1996 8.09 3600 0.3627 0.4660
0.1755 8.99 4000 0.3582 0.4619
0.1697 9.89 4400 0.3581 0.4555

Framework versions

  • Transformers 4.6.0
  • Pytorch 1.8.1+cu102
  • Datasets 1.6.2
  • Tokenizers 0.10.2