Edit model card

t5-base-TEDxJP-0front-1body-10rear-order-RB

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: 0.4749
  • Wer: 0.1754
  • Mer: 0.1696
  • Wil: 0.2575
  • Wip: 0.7425
  • Hits: 55482
  • Substitutions: 6478
  • Deletions: 2627
  • Insertions: 2225
  • Cer: 0.1370

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: 32
  • eval_batch_size: 32
  • seed: 40
  • 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
0.637 1.0 1457 0.4932 0.2359 0.2179 0.3082 0.6918 54682 6909 2996 5331 0.2100
0.5501 2.0 2914 0.4572 0.1831 0.1766 0.2655 0.7345 55134 6575 2878 2370 0.1461
0.5505 3.0 4371 0.4470 0.1787 0.1728 0.2609 0.7391 55267 6494 2826 2222 0.1400
0.4921 4.0 5828 0.4426 0.1794 0.1730 0.2606 0.7394 55420 6468 2699 2423 0.1407
0.4465 5.0 7285 0.4507 0.1783 0.1721 0.2596 0.7404 55420 6458 2709 2351 0.1390
0.3557 6.0 8742 0.4567 0.1768 0.1708 0.2585 0.7415 55416 6459 2712 2245 0.1401
0.3367 7.0 10199 0.4613 0.1772 0.1709 0.2589 0.7411 55505 6497 2585 2363 0.1387
0.328 8.0 11656 0.4624 0.1769 0.1708 0.2587 0.7413 55442 6478 2667 2278 0.1383
0.2992 9.0 13113 0.4726 0.1764 0.1704 0.2580 0.7420 55461 6463 2663 2264 0.1378
0.2925 10.0 14570 0.4749 0.1754 0.1696 0.2575 0.7425 55482 6478 2627 2225 0.1370

Framework versions

  • Transformers 4.21.2
  • Pytorch 1.12.1+cu116
  • Datasets 2.4.0
  • Tokenizers 0.12.1
Downloads last month
23
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.