Hugging Face's logo

NbAiLab
/
nb-wav2vec2-1b-nynorsk

Automatic Speech Recognition
Transformers
PyTorch
TensorBoard
Norwegian Nynorsk
Norwegian
wav2vec2
NbAiLab/NPSC
nb-NN
Eval Results
Inference Endpoints
Model card Files Metrics Community
nb-wav2vec2-1b-nynorsk / README.old.md
versae's picture
versae
Create README.old.md
6147168
preview code
|
raw
history blame
5.38 kB
metadata
license: apache-2.0
tags:
  - automatic-speech-recognition
  - NbAiLab/NPSC
  - generated_from_trainer
model-index:
  - name: XLSR-1B-nynorsk-low
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: NPSC
          type: NbAiLab/NPSC
          args: 16K_mp3_nynorsk
        metrics:
          - name: Test (Nynorsk) WER
            type: wer
            value: 0.11319692134409612
          - name: Test (Nynorsk) CER
            type: cer
            value: 0.040263696587740365

XLSR-1B-nynorsk-low

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the NBAILAB/NPSC - 16K_MP3_NYNORSK dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2909
  • Wer: 0.1364

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: 2e-05
  • train_batch_size: 24
  • eval_batch_size: 24
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 2000
  • num_epochs: 60.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
2.8979 1.0 500 2.9413 1.0
1.2224 2.0 1000 1.0359 0.7802
0.8643 3.01 1500 0.7746 0.5969
0.8211 4.01 2000 0.4882 0.3710
0.5287 5.01 2500 0.4060 0.3085
0.4724 6.01 3000 0.3297 0.2517
0.4357 7.01 3500 0.3106 0.2342
0.376 8.02 4000 0.2776 0.2072
0.3286 9.02 4500 0.2888 0.2032
0.3731 10.02 5000 0.2691 0.1835
0.306 11.02 5500 0.2536 0.1835
0.3025 12.02 6000 0.2758 0.1809
0.3413 13.03 6500 0.2791 0.1823
0.2601 14.03 7000 0.2912 0.1759
0.2332 15.03 7500 0.2582 0.1694
0.2108 16.03 8000 0.2717 0.1660
0.2122 17.03 8500 0.2848 0.1647
0.2369 18.04 9000 0.2548 0.1646
0.1906 19.04 9500 0.2667 0.1627
0.1943 20.04 10000 0.2662 0.1623
0.18 21.04 10500 0.2769 0.1561
0.1654 22.04 11000 0.2661 0.1558
0.1515 23.05 11500 0.2870 0.1597
0.147 24.05 12000 0.2778 0.1551
0.1622 25.05 12500 0.2753 0.1541
0.1522 26.05 13000 0.2932 0.1521
0.1522 27.05 13500 0.2548 0.1513
0.1319 28.06 14000 0.2811 0.1532
0.1261 29.06 14500 0.2786 0.1521
0.1391 30.06 15000 0.2651 0.1461
0.1486 31.06 15500 0.2866 0.1494
0.1121 32.06 16000 0.2641 0.1478
0.1114 33.07 16500 0.2910 0.1478
0.101 34.07 17000 0.2884 0.1443
0.1135 35.07 17500 0.3029 0.1469
0.0972 36.07 18000 0.2870 0.1467
0.1178 37.07 18500 0.2745 0.1450
0.0885 38.08 19000 0.2836 0.1440
0.1144 39.08 19500 0.2761 0.1446
0.0997 40.08 20000 0.2806 0.1439
0.1012 41.08 20500 0.2878 0.1413
0.0902 42.08 21000 0.2832 0.1452
0.0804 43.09 21500 0.2911 0.1458
0.0762 44.09 22000 0.2708 0.1441
0.0758 45.09 22500 0.2804 0.1434
0.0874 46.09 23000 0.2831 0.1407
0.0895 47.09 23500 0.2913 0.1396
0.0975 48.1 24000 0.2956 0.1411
0.0758 49.1 24500 0.2920 0.1385
0.0704 50.1 25000 0.2788 0.1383
0.0707 51.1 25500 0.2822 0.1388
0.0664 52.1 26000 0.2876 0.1371
0.0692 53.11 26500 0.2815 0.1377
0.0799 54.11 27000 0.2806 0.1363
0.0611 55.11 27500 0.2878 0.1363
0.0759 56.11 28000 0.2900 0.1365
0.0801 57.11 28500 0.2881 0.1375
0.0644 58.12 29000 0.2898 0.1362
0.068 59.12 29500 0.2913 0.1369

Framework versions

  • Transformers 4.17.0.dev0
  • Pytorch 1.10.2+cu113
  • Datasets 2.0.0
  • Tokenizers 0.11.0