Edit model card

wav2vec2-xls-r-300m-eo

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

  • Loss: 0.2584
  • Wer: 0.3114

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
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 20.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
3.1701 0.8 500 2.8105 1.0
1.9143 1.6 1000 0.5977 0.7002
1.1259 2.4 1500 0.5063 0.6157
0.9732 3.2 2000 0.4264 0.5673
0.8983 4.0 2500 0.4249 0.4902
0.8507 4.8 3000 0.3811 0.4536
0.8064 5.6 3500 0.3643 0.4467
0.7866 6.4 4000 0.3600 0.4453
0.7773 7.2 4500 0.3724 0.4470
0.747 8.0 5000 0.3501 0.4189
0.7279 8.8 5500 0.3500 0.4261
0.7153 9.6 6000 0.3328 0.3966
0.7 10.4 6500 0.3314 0.3869
0.6784 11.2 7000 0.3396 0.4051
0.6582 12.0 7500 0.3236 0.3899
0.6478 12.8 8000 0.3263 0.3832
0.6277 13.6 8500 0.3139 0.3769
0.6053 14.4 9000 0.2955 0.3536
0.5777 15.2 9500 0.2793 0.3413
0.5631 16.0 10000 0.2789 0.3353
0.5446 16.8 10500 0.2709 0.3264
0.528 17.6 11000 0.2693 0.3234
0.5169 18.4 11500 0.2656 0.3193
0.5041 19.2 12000 0.2575 0.3102
0.4971 20.0 12500 0.2584 0.3114

Framework versions

  • Transformers 4.17.0.dev0
  • Pytorch 1.10.2+cu102
  • Datasets 1.18.2.dev0
  • Tokenizers 0.11.0

Evaluation Commands

  1. To evaluate on mozilla-foundation/common_voice_7_0 with split test
python eval.py --model_id samitizerxu/wav2vec2-xls-r-300m-eo --dataset mozilla-foundation/common_voice_7_0 --config eo --split test
Downloads last month
12

Dataset used to train samitizerxu/wav2vec2-xls-r-300m-eo

Evaluation results