This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_9_0 - BN dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2297
  • Wer: 0.2850
  • Cer: 0.0660

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: 7.5e-05
  • train_batch_size: 64
  • eval_batch_size: 64
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 128
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • training_steps: 8692
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer Cer
3.675 2.3 400 3.5052 1.0 1.0
3.0446 4.6 800 2.2759 1.0052 0.5215
1.7276 6.9 1200 0.7083 0.6697 0.1969
1.5171 9.2 1600 0.5328 0.5733 0.1568
1.4176 11.49 2000 0.4571 0.5161 0.1381
1.343 13.79 2400 0.3910 0.4522 0.1160
1.2743 16.09 2800 0.3534 0.4137 0.1044
1.2396 18.39 3200 0.3278 0.3877 0.0959
1.2035 20.69 3600 0.3109 0.3741 0.0917
1.1745 22.99 4000 0.2972 0.3618 0.0882
1.1541 25.29 4400 0.2836 0.3427 0.0832
1.1372 27.59 4800 0.2759 0.3357 0.0812
1.1048 29.89 5200 0.2669 0.3284 0.0783
1.0966 32.18 5600 0.2678 0.3249 0.0775
1.0747 34.48 6000 0.2547 0.3134 0.0748
1.0593 36.78 6400 0.2491 0.3077 0.0728
1.0417 39.08 6800 0.2450 0.3012 0.0711
1.024 41.38 7200 0.2402 0.2956 0.0694
1.0106 43.68 7600 0.2351 0.2915 0.0681
1.0014 45.98 8000 0.2328 0.2896 0.0673
0.9999 48.28 8400 0.2318 0.2866 0.0667

Framework versions

  • Transformers 4.19.0.dev0
  • Pytorch 1.11.0+cu102
  • Datasets 2.1.1.dev0
  • Tokenizers 0.12.1
Downloads last month
30
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train anuragshas/wav2vec2-xls-r-300m-bn-cv9-with-lm

Evaluation results