Hugging Face's logo

utakumi
/
Hubert-common_voice-ja-demo-kana-only-cosine

Automatic Speech Recognition
Transformers
Safetensors
Japanese
hubert
mozilla-foundation/common_voice_13_0
Generated from Trainer
Eval Results
Inference Endpoints
Model card Files Community
Hubert-common_voice-ja-demo-kana-only-cosine / README.md
utakumi's picture
utakumi
End of training
4f6ba30 verified
preview code
|
raw
history blame contribute delete
8.81 kB
metadata
library_name: transformers
language:
  - ja
license: apache-2.0
base_model: rinna/japanese-hubert-base
tags:
  - automatic-speech-recognition
  - mozilla-foundation/common_voice_13_0
  - generated_from_trainer
datasets:
  - common_voice_13_0
metrics:
  - wer
model-index:
  - name: Hubert-common_voice-ja-demo-kana-only-cosine
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: MOZILLA-FOUNDATION/COMMON_VOICE_13_0 - JA
          type: common_voice_13_0
          config: ja
          split: test
          args: 'Config: ja, Training split: train+validation, Eval split: test'
        metrics:
          - name: Wer
            type: wer
            value: 1

Hubert-common_voice-ja-demo-kana-only-cosine

This model is a fine-tuned version of rinna/japanese-hubert-base on the MOZILLA-FOUNDATION/COMMON_VOICE_13_0 - JA dataset. It achieves the following results on the evaluation set:

  • Loss: 0.6322
  • Wer: 1.0
  • Cer: 0.3308

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: 3e-05
  • train_batch_size: 16
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 32
  • optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 12500
  • num_epochs: 25.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer Cer
No log 0.2660 100 43.2251 1.5295 5.8210
No log 0.5319 200 42.4452 1.5322 5.2849
No log 0.7979 300 40.4154 1.1262 1.8126
No log 1.0638 400 32.8021 1.0 0.9999
31.884 1.3298 500 20.8133 1.0 0.9999
31.884 1.5957 600 17.5824 1.0 0.9999
31.884 1.8617 700 16.8682 1.0 0.9999
31.884 2.1277 800 16.4466 1.0 0.9999
31.884 2.3936 900 16.0037 1.0 0.9999
14.4701 2.6596 1000 15.5409 1.0 0.9999
14.4701 2.9255 1100 15.0446 1.0 0.9999
14.4701 3.1915 1200 14.5023 1.0 0.9999
14.4701 3.4574 1300 13.9298 1.0 0.9999
14.4701 3.7234 1400 13.3212 1.0 0.9999
12.1626 3.9894 1500 12.6814 1.0 0.9999
12.1626 4.2553 1600 12.0099 1.0 0.9999
12.1626 4.5213 1700 11.3179 1.0 0.9999
12.1626 4.7872 1800 10.6017 1.0 0.9999
12.1626 5.0532 1900 9.8810 1.0 0.9999
9.5127 5.3191 2000 9.1567 1.0 0.9999
9.5127 5.5851 2100 8.4445 1.0 0.9999
9.5127 5.8511 2200 7.7573 1.0 0.9999
9.5127 6.1170 2300 7.1049 1.0 0.9999
9.5127 6.3830 2400 6.5016 1.0 0.9999
6.6873 6.6489 2500 5.9565 1.0 0.9999
6.6873 6.9149 2600 5.4853 1.0 0.9999
6.6873 7.1809 2700 5.0997 1.0 0.9999
6.6873 7.4468 2800 4.7889 1.0 0.9999
6.6873 7.7128 2900 4.5573 1.0 0.9999
4.7448 7.9787 3000 4.3889 1.0 0.9999
4.7448 8.2447 3100 4.2614 1.0 0.9999
4.7448 8.5106 3200 4.1960 1.0 0.9999
4.7448 8.7766 3300 4.1398 1.0 0.9999
4.7448 9.0426 3400 4.1092 1.0 0.9999
4.1253 9.3085 3500 4.0911 1.0 0.9999
4.1253 9.5745 3600 4.0851 1.0 0.9999
4.1253 9.8404 3700 4.0707 1.0 0.9999
4.1253 10.1064 3800 4.0630 1.0 0.9999
4.1253 10.3723 3900 4.0589 1.0 0.9999
4.0399 10.6383 4000 4.0574 1.0 0.9999
4.0399 10.9043 4100 4.0495 1.0 0.9999
4.0399 11.1702 4200 4.0367 1.0 0.9999
4.0399 11.4362 4300 4.0297 1.0 0.9999
4.0399 11.7021 4400 4.0168 1.0 0.9999
4.0102 11.9681 4500 4.0002 1.0 0.9999
4.0102 12.2340 4600 3.9823 1.0 0.9999
4.0102 12.5 4700 3.9474 1.0 0.9999
4.0102 12.7660 4800 3.8870 1.0 0.9999
4.0102 13.0319 4900 3.7933 1.0 0.9999
3.8616 13.2979 5000 3.6576 1.0 0.9999
3.8616 13.5638 5100 3.4925 1.0 0.9999
3.8616 13.8298 5200 3.2550 1.0 0.9999
3.8616 14.0957 5300 2.8836 1.0 0.8301
3.8616 14.3617 5400 2.5211 1.0 0.6171
3.023 14.6277 5500 2.2902 1.0 0.5481
3.023 14.8936 5600 2.1006 1.0 0.5079
3.023 15.1596 5700 1.9464 1.0 0.4784
3.023 15.4255 5800 1.8196 1.0 0.4597
3.023 15.6915 5900 1.6975 1.0 0.4238
1.9348 15.9574 6000 1.6040 1.0 0.4093
1.9348 16.2234 6100 1.5035 1.0 0.4021
1.9348 16.4894 6200 1.4211 1.0 0.3930
1.9348 16.7553 6300 1.3529 1.0 0.3802
1.9348 17.0213 6400 1.2795 1.0 0.3791
1.4128 17.2872 6500 1.2193 1.0 0.3711
1.4128 17.5532 6600 1.1646 1.0 0.3674
1.4128 17.8191 6700 1.1193 1.0 0.3706
1.4128 18.0851 6800 1.0665 1.0 0.3606
1.4128 18.3511 6900 1.0244 0.9998 0.3590
1.1012 18.6170 7000 0.9864 1.0 0.3540
1.1012 18.8830 7100 0.9578 1.0 0.3554
1.1012 19.1489 7200 0.9309 0.9998 0.3509
1.1012 19.4149 7300 0.9070 1.0 0.3495
1.1012 19.6809 7400 0.8693 0.9998 0.3470
0.9083 19.9468 7500 0.8492 1.0 0.3449
0.9083 20.2128 7600 0.8214 1.0 0.3449
0.9083 20.4787 7700 0.8211 1.0 0.3500
0.9083 20.7447 7800 0.7964 1.0 0.3452
0.9083 21.0106 7900 0.7797 1.0 0.3429
0.7546 21.2766 8000 0.7634 1.0 0.3400
0.7546 21.5426 8100 0.7471 1.0 0.3384
0.7546 21.8085 8200 0.7400 1.0 0.3378
0.7546 22.0745 8300 0.7214 1.0 0.3390
0.7546 22.3404 8400 0.7062 0.9998 0.3375
0.651 22.6064 8500 0.6973 1.0 0.3344
0.651 22.8723 8600 0.6930 0.9998 0.3344
0.651 23.1383 8700 0.6829 1.0 0.3350
0.651 23.4043 8800 0.6683 1.0 0.3332
0.651 23.6702 8900 0.6596 0.9998 0.3322
0.5868 23.9362 9000 0.6764 1.0 0.3321
0.5868 24.2021 9100 0.6635 0.9998 0.3308
0.5868 24.4681 9200 0.6560 1.0 0.3324
0.5868 24.7340 9300 0.6412 1.0 0.3290
0.5868 25.0 9400 0.6323 1.0 0.3307

Framework versions

  • Transformers 4.47.0.dev0
  • Pytorch 2.5.1+cu124
  • Datasets 3.1.0
  • Tokenizers 0.20.3