Edit model card

t5-base-TEDxJP-11body-0context

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: 0.8068
  • Wer: 0.1976
  • Mer: 0.1904
  • Wil: 0.2816
  • Wip: 0.7184
  • Hits: 602335
  • Substitutions: 75050
  • Deletions: 39435
  • Insertions: 27185
  • Cer: 0.1625

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
0.8909 1.0 746 0.7722 0.3120 0.2861 0.3989 0.6011 558138 99887 58795 64983 0.2652
0.6786 2.0 1492 0.7021 0.2226 0.2122 0.3069 0.6931 592242 78773 45805 34978 0.1862
0.5627 3.0 2238 0.6996 0.2104 0.2016 0.2942 0.7058 597381 76593 42846 31392 0.1752
0.489 4.0 2984 0.7161 0.2030 0.1952 0.2865 0.7135 599808 75155 41857 28506 0.1684
0.4355 5.0 3730 0.7389 0.2000 0.1924 0.2837 0.7163 601815 75247 39758 28335 0.1651
0.3836 6.0 4476 0.7537 0.1992 0.1918 0.2829 0.7171 601846 75046 39928 27815 0.1640
0.3617 7.0 5222 0.7743 0.1995 0.1918 0.2832 0.7168 602287 75268 39265 28445 0.1642
0.3258 8.0 5968 0.7907 0.1971 0.1899 0.2809 0.7191 602800 74887 39133 27258 0.1620
0.3225 9.0 6714 0.8035 0.1981 0.1908 0.2823 0.7177 602418 75372 39030 27625 0.1630
0.3162 10.0 7460 0.8068 0.1976 0.1904 0.2816 0.7184 602335 75050 39435 27185 0.1625

Framework versions

  • Transformers 4.12.5
  • Pytorch 1.10.0+cu102
  • Datasets 1.15.1
  • Tokenizers 0.10.3
Downloads last month
7
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.