Edit model card

t5-base-TEDxJP-1body-1context

This model is a fine-tuned version of sonoisa/t5-base-japanese on the te_dx_jp dataset. It achieves the following results on the evaluation set:

  • Loss: 0.5061
  • Wer: 0.1990
  • Mer: 0.1913
  • Wil: 0.2823
  • Wip: 0.7177
  • Hits: 55830
  • Substitutions: 6943
  • Deletions: 3598
  • Insertions: 2664
  • Cer: 0.1763

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: 64
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 10

Training results

Training Loss Epoch Step Validation Loss Wer Mer Wil Wip Hits Substitutions Deletions Insertions Cer
0.7277 1.0 746 0.5799 0.2384 0.2256 0.3188 0.6812 54323 7170 4878 3777 0.2371
0.6278 2.0 1492 0.5254 0.2070 0.1997 0.2905 0.7095 55045 6885 4441 2412 0.1962
0.5411 3.0 2238 0.5076 0.2022 0.1950 0.2858 0.7142 55413 6902 4056 2463 0.1805
0.53 4.0 2984 0.5020 0.1979 0.1911 0.2814 0.7186 55599 6849 3923 2362 0.1761
0.5094 5.0 3730 0.4999 0.1987 0.1915 0.2828 0.7172 55651 6944 3776 2465 0.1742
0.4783 6.0 4476 0.5016 0.1985 0.1914 0.2826 0.7174 55684 6947 3740 2490 0.1753
0.4479 7.0 5222 0.5035 0.1976 0.1905 0.2819 0.7181 55726 6961 3684 2468 0.1733
0.4539 8.0 5968 0.5022 0.1967 0.1896 0.2807 0.7193 55795 6938 3638 2477 0.1729
0.4632 9.0 6714 0.5034 0.1991 0.1913 0.2824 0.7176 55844 6942 3585 2687 0.1758
0.4201 10.0 7460 0.5061 0.1990 0.1913 0.2823 0.7177 55830 6943 3598 2664 0.1763

Framework versions

  • Transformers 4.12.5
  • Pytorch 1.10.0+cu102
  • Datasets 1.15.1
  • Tokenizers 0.10.3
Downloads last month
16
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.