wav2vec2-large-xlsr-hindi-demo-colab_2

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice dataset. It achieves the following results on the evaluation set:

  • Loss: 3.8793
  • Wer: 1.1357

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: 8
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 16
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 20
  • num_epochs: 50
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
22.381 1.11 20 22.1964 1.0
7.6212 2.22 40 4.0591 1.0
3.6951 3.32 60 3.6782 1.0
3.5574 4.43 80 3.6776 1.0
3.5374 5.54 100 3.5649 1.0
3.5512 6.65 120 3.5266 1.0
3.5075 7.76 140 3.6860 1.0
3.5097 8.86 160 3.4941 1.0
3.481 9.97 180 3.4659 1.0
3.5623 11.11 200 3.7254 1.0
3.4404 12.22 220 3.5225 1.0
3.432 13.32 240 3.5706 1.0
3.4177 14.43 260 3.3833 1.0
3.3735 15.54 280 3.4140 1.0
3.31 16.65 300 3.2702 1.0
3.2256 17.76 320 3.2405 1.0
3.0546 18.86 340 3.1644 1.0
2.7233 19.97 360 2.9753 1.0
2.2822 21.11 380 3.1119 1.1183
1.8027 22.22 400 3.0035 1.2378
1.5274 23.32 420 2.8536 1.2227
1.2313 24.43 440 2.9544 1.0951
1.0956 25.54 460 2.8814 1.0661
0.9456 26.65 480 3.1192 1.1589
0.7893 27.76 500 3.2919 1.1833
0.7256 28.86 520 3.0864 1.0951
0.6051 29.97 540 3.5888 1.1821
0.6087 31.11 560 3.4579 1.1392
0.5529 32.22 580 3.1998 1.0708
0.5211 33.32 600 3.4655 1.1311
0.4506 34.43 620 3.4338 1.1694
0.4101 35.54 640 3.5189 1.1450
0.4484 36.65 660 3.6585 1.1601
0.4038 37.76 680 3.6314 1.1497
0.3539 38.86 700 3.6955 1.1485
0.3898 39.97 720 3.5738 1.1148
0.35 41.11 740 3.6594 1.1195
0.3328 42.22 760 3.6894 1.1299
0.3264 43.32 780 3.7290 1.1021
0.3364 44.43 800 3.7256 1.1543
0.3071 45.54 820 3.8834 1.1415
0.3074 46.65 840 3.8077 1.1450
0.3064 47.76 860 3.8733 1.1346
0.3223 48.86 880 3.8780 1.1323
0.275 49.97 900 3.8793 1.1357

Framework versions

  • Transformers 4.11.3
  • Pytorch 1.10.0+cu111
  • Datasets 1.13.3
  • Tokenizers 0.10.3
Downloads last month
1
Hosted inference API
or or
This model can be loaded on the Inference API on-demand.

Dataset used to train rafiulrumy/wav2vec2-large-xlsr-hindi-demo-colab_2

Evaluation results