Hubert-common_voice-phoneme-ctc_zero_infinity

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: 0.5230
  • Wer: 1.0
  • Cer: 0.1953

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: 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: 20.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer Cer
No log 0.2660 100 18.2381 1.1471 1.8153
No log 0.5319 200 8.1726 1.0 0.9817
No log 0.7979 300 6.9386 1.0 0.9817
No log 1.0638 400 6.2389 1.0 0.9817
8.8178 1.3298 500 5.4653 1.0 0.9817
8.8178 1.5957 600 4.6745 1.0 0.9817
8.8178 1.8617 700 3.9771 1.0 0.9817
8.8178 2.1277 800 3.4579 1.0 0.9817
8.8178 2.3936 900 3.1745 1.0 0.9817
3.6858 2.6596 1000 3.0675 1.0 0.9817
3.6858 2.9255 1100 3.0343 1.0 0.9817
3.6858 3.1915 1200 3.0102 1.0 0.9817
3.6858 3.4574 1300 2.9925 1.0 0.9817
3.6858 3.7234 1400 2.5595 1.0 0.9367
2.7891 3.9894 1500 1.5432 1.0 0.3742
2.7891 4.2553 1600 1.0799 1.0 0.2972
2.7891 4.5213 1700 0.8670 1.0 0.2639
2.7891 4.7872 1800 0.7350 1.0 0.2559
2.7891 5.0532 1900 0.6753 1.0 0.2468
0.9179 5.3191 2000 0.6171 1.0 0.2389
0.9179 5.5851 2100 0.5866 1.0 0.2386
0.9179 5.8511 2200 0.5649 1.0 0.2389
0.9179 6.1170 2300 0.5368 1.0 0.2321
0.9179 6.3830 2400 0.5225 1.0 0.2289
0.563 6.6489 2500 0.5042 1.0 0.2293
0.563 6.9149 2600 0.4918 1.0 0.2247
0.563 7.1809 2700 0.4881 1.0 0.2208
0.563 7.4468 2800 0.4787 1.0 0.2198
0.563 7.7128 2900 0.4692 1.0 0.2181
0.4453 7.9787 3000 0.4733 1.0 0.2151
0.4453 8.2447 3100 0.4585 1.0 0.2147
0.4453 8.5106 3200 0.4463 1.0 0.2116
0.4453 8.7766 3300 0.4183 1.0 0.2055
0.4453 9.0426 3400 0.4308 0.9998 0.2032
0.3596 9.3085 3500 0.4070 1.0 0.2022
0.3596 9.5745 3600 0.4259 1.0 0.2024
0.3596 9.8404 3700 0.4038 1.0 0.1985
0.3596 10.1064 3800 0.4272 1.0 0.1976
0.3596 10.3723 3900 0.3961 0.9998 0.1969
0.2945 10.6383 4000 0.4180 1.0 0.1943
0.2945 10.9043 4100 0.3999 1.0 0.1975
0.2945 11.1702 4200 0.3879 1.0 0.1930
0.2945 11.4362 4300 0.3799 1.0 0.1918
0.2945 11.7021 4400 0.3764 0.9998 0.1927
0.2605 11.9681 4500 0.3725 1.0 0.1919
0.2605 12.2340 4600 0.3910 1.0 0.1919
0.2605 12.5 4700 0.3851 0.9996 0.1908
0.2605 12.7660 4800 0.4115 1.0 0.1906
0.2605 13.0319 4900 0.3779 1.0 0.1894
0.2223 13.2979 5000 0.3956 1.0 0.1904
0.2223 13.5638 5100 0.4001 1.0 0.1907
0.2223 13.8298 5200 0.3891 1.0 0.1948
0.2223 14.0957 5300 0.3940 1.0 0.1902
0.2223 14.3617 5400 0.4056 1.0 0.1909
0.211 14.6277 5500 0.4000 0.9998 0.1929
0.211 14.8936 5600 0.3926 1.0 0.1895
0.211 15.1596 5700 0.3852 0.9998 0.1930
0.211 15.4255 5800 0.3864 1.0 0.1886
0.211 15.6915 5900 0.3951 0.9998 0.1909
0.1983 15.9574 6000 0.3951 1.0 0.1882
0.1983 16.2234 6100 0.4087 1.0 0.1918
0.1983 16.4894 6200 0.4150 1.0 0.1891
0.1983 16.7553 6300 0.4008 0.9998 0.1907
0.1983 17.0213 6400 0.4220 1.0 0.1943
0.1829 17.2872 6500 0.4154 1.0 0.1925
0.1829 17.5532 6600 0.4482 1.0 0.1959
0.1829 17.8191 6700 0.4217 0.9998 0.1939
0.1829 18.0851 6800 0.4383 0.9998 0.1916
0.1829 18.3511 6900 0.4226 1.0 0.1926
0.1757 18.6170 7000 0.4170 0.9998 0.1916
0.1757 18.8830 7100 0.4162 1.0 0.1918
0.1757 19.1489 7200 0.4350 0.9998 0.1910
0.1757 19.4149 7300 0.4403 1.0 0.2022
0.1757 19.6809 7400 0.4325 0.9998 0.1944
0.1801 19.9468 7500 0.5488 1.0 0.1977

Framework versions

  • Transformers 4.47.0.dev0
  • Pytorch 2.5.1+cu124
  • Datasets 3.1.0
  • Tokenizers 0.20.3
Downloads last month
31
Safetensors
Model size
94.4M params
Tensor type
F32
·
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.

Model tree for utakumi/Hubert-common_voice-phoneme-ctc_zero_infinity

Finetuned
(21)
this model

Evaluation results