Edit model card

wav2vec2-xls-r-gn-cv7

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

  • Loss: 1.7197
  • Wer: 0.7434

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: 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: 100
  • training_steps: 13000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
3.4669 6.24 100 3.3003 1.0
3.3214 12.48 200 3.2090 1.0
3.1619 18.73 300 2.6322 1.0
1.751 24.97 400 1.4089 0.9803
0.7997 31.24 500 0.9996 0.9211
0.4996 37.48 600 0.9879 0.8553
0.3677 43.73 700 0.9543 0.8289
0.2851 49.97 800 1.0627 0.8487
0.2556 56.24 900 1.0933 0.8355
0.2268 62.48 1000 0.9191 0.8026
0.1914 68.73 1100 0.9582 0.7961
0.1749 74.97 1200 1.0502 0.8092
0.157 81.24 1300 0.9998 0.7632
0.1505 87.48 1400 1.0076 0.7303
0.1278 93.73 1500 0.9321 0.75
0.1078 99.97 1600 1.0383 0.7697
0.1156 106.24 1700 1.0302 0.7763
0.1107 112.48 1800 1.0419 0.7763
0.091 118.73 1900 1.0694 0.75
0.0829 124.97 2000 1.0257 0.7829
0.0865 131.24 2100 1.2108 0.7368
0.0907 137.48 2200 1.0458 0.7697
0.0897 143.73 2300 1.1504 0.7895
0.0766 149.97 2400 1.1663 0.7237
0.0659 156.24 2500 1.1320 0.7632
0.0699 162.48 2600 1.2586 0.7434
0.0613 168.73 2700 1.1815 0.8158
0.0598 174.97 2800 1.3299 0.75
0.0577 181.24 2900 1.2035 0.7171
0.0576 187.48 3000 1.2134 0.7434
0.0518 193.73 3100 1.3406 0.7566
0.0524 199.97 3200 1.4251 0.75
0.0467 206.24 3300 1.3533 0.7697
0.0428 212.48 3400 1.2463 0.7368
0.0453 218.73 3500 1.4532 0.7566
0.0473 224.97 3600 1.3152 0.7434
0.0451 231.24 3700 1.2232 0.7368
0.0361 237.48 3800 1.2938 0.7171
0.045 243.73 3900 1.4148 0.7434
0.0422 249.97 4000 1.3786 0.7961
0.036 256.24 4100 1.4488 0.7697
0.0352 262.48 4200 1.2294 0.6776
0.0326 268.73 4300 1.2796 0.6974
0.034 274.97 4400 1.3805 0.7303
0.0305 281.24 4500 1.4994 0.7237
0.0325 287.48 4600 1.4330 0.6908
0.0338 293.73 4700 1.3091 0.7368
0.0306 299.97 4800 1.2174 0.7171
0.0299 306.24 4900 1.3527 0.7763
0.0287 312.48 5000 1.3651 0.7368
0.0274 318.73 5100 1.4337 0.7368
0.0258 324.97 5200 1.3831 0.6908
0.022 331.24 5300 1.3556 0.6974
0.021 337.48 5400 1.3836 0.7237
0.0241 343.73 5500 1.4352 0.7039
0.0229 349.97 5600 1.3904 0.7105
0.026 356.24 5700 1.4131 0.7171
0.021 362.48 5800 1.5426 0.6974
0.0191 368.73 5900 1.5960 0.7632
0.0227 374.97 6000 1.6240 0.7368
0.0204 381.24 6100 1.4301 0.7105
0.0175 387.48 6200 1.5554 0.75
0.0183 393.73 6300 1.6044 0.7697
0.0183 399.97 6400 1.5963 0.7368
0.016 406.24 6500 1.5679 0.7829
0.0178 412.48 6600 1.5928 0.7697
0.014 418.73 6700 1.7000 0.7632
0.0182 424.97 6800 1.5340 0.75
0.0148 431.24 6900 1.9274 0.7368
0.0148 437.48 7000 1.6437 0.7697
0.0173 443.73 7100 1.5468 0.75
0.0109 449.97 7200 1.6083 0.75
0.0167 456.24 7300 1.6732 0.75
0.0139 462.48 7400 1.5097 0.7237
0.013 468.73 7500 1.5947 0.7171
0.0128 474.97 7600 1.6260 0.7105
0.0166 481.24 7700 1.5756 0.7237
0.0127 487.48 7800 1.4506 0.6908
0.013 493.73 7900 1.4882 0.7368
0.0125 499.97 8000 1.5589 0.7829
0.0141 506.24 8100 1.6328 0.7434
0.0115 512.48 8200 1.6586 0.7434
0.0117 518.73 8300 1.6043 0.7105
0.009 524.97 8400 1.6508 0.7237
0.0108 531.24 8500 1.4507 0.6974
0.011 537.48 8600 1.5942 0.7434
0.009 543.73 8700 1.8121 0.7697
0.0112 549.97 8800 1.6923 0.7697
0.0073 556.24 8900 1.7096 0.7368
0.0098 562.48 9000 1.7052 0.7829
0.0088 568.73 9100 1.6956 0.7566
0.0099 574.97 9200 1.4909 0.7171
0.0075 581.24 9300 1.6307 0.7697
0.0077 587.48 9400 1.6196 0.7961
0.0088 593.73 9500 1.6119 0.7566
0.0085 599.97 9600 1.4512 0.7368
0.0086 606.24 9700 1.5992 0.7237
0.0109 612.48 9800 1.4706 0.7368
0.0098 618.73 9900 1.3824 0.7171
0.0091 624.97 10000 1.4776 0.6974
0.0072 631.24 10100 1.4896 0.7039
0.0087 637.48 10200 1.5467 0.7368
0.007 643.73 10300 1.5493 0.75
0.0076 649.97 10400 1.5706 0.7303
0.0085 656.24 10500 1.5748 0.7237
0.0075 662.48 10600 1.5081 0.7105
0.0068 668.73 10700 1.4967 0.6842
0.0117 674.97 10800 1.4986 0.7105
0.0054 681.24 10900 1.5587 0.7303
0.0059 687.48 11000 1.5886 0.7171
0.0071 693.73 11100 1.5746 0.7171
0.0048 699.97 11200 1.6166 0.7237
0.0048 706.24 11300 1.6098 0.7237
0.0056 712.48 11400 1.5834 0.7237
0.0048 718.73 11500 1.5653 0.7171
0.0045 724.97 11600 1.6252 0.7237
0.0068 731.24 11700 1.6794 0.7171
0.0044 737.48 11800 1.6881 0.7039
0.008 743.73 11900 1.7393 0.75
0.0045 749.97 12000 1.6869 0.7237
0.0047 756.24 12100 1.7105 0.7303
0.0057 762.48 12200 1.7439 0.7303
0.004 768.73 12300 1.7871 0.7434
0.0061 774.97 12400 1.7812 0.7303
0.005 781.24 12500 1.7410 0.7434
0.0056 787.48 12600 1.7220 0.7303
0.0064 793.73 12700 1.7141 0.7434
0.0042 799.97 12800 1.7139 0.7368
0.0049 806.24 12900 1.7211 0.7434
0.0044 812.48 13000 1.7197 0.7434

Framework versions

  • Transformers 4.15.0
  • Pytorch 1.10.0+cu111
  • Datasets 1.18.0
  • 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 lgris/wav2vec2-xls-r-gn-cv7

Evaluation results