<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
t5-base-TEDxJP-1front-1body-0rear
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: 0.4787
- Wer: 0.1786
- Mer: 0.1722
- Wil: 0.2608
- Wip: 0.7392
- Hits: 55434
- Substitutions: 6554
- Deletions: 2599
- Insertions: 2380
- Cer: 0.1399
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: 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6606 | 1.0 | 1457 | 0.5113 | 0.2142 | 0.2017 | 0.2939 | 0.7061 | 54751 | 6976 | 2860 | 4000 | 0.1909 |
| 0.5636 | 2.0 | 2914 | 0.4669 | 0.1913 | 0.1832 | 0.2728 | 0.7272 | 55086 | 6669 | 2832 | 2852 | 0.1700 |
| 0.5115 | 3.0 | 4371 | 0.4543 | 0.1815 | 0.1747 | 0.2633 | 0.7367 | 55384 | 6559 | 2644 | 2519 | 0.1504 |
| 0.4463 | 4.0 | 5828 | 0.4512 | 0.1796 | 0.1733 | 0.2617 | 0.7383 | 55344 | 6534 | 2709 | 2358 | 0.1422 |
| 0.4001 | 5.0 | 7285 | 0.4564 | 0.1779 | 0.1718 | 0.2600 | 0.7400 | 55394 | 6509 | 2684 | 2295 | 0.1395 |
| 0.3683 | 6.0 | 8742 | 0.4600 | 0.1790 | 0.1726 | 0.2611 | 0.7389 | 55436 | 6546 | 2605 | 2413 | 0.1405 |
| 0.391 | 7.0 | 10199 | 0.4651 | 0.1781 | 0.1718 | 0.2599 | 0.7401 | 55424 | 6505 | 2658 | 2338 | 0.1391 |
| 0.337 | 8.0 | 11656 | 0.4705 | 0.1775 | 0.1714 | 0.2595 | 0.7405 | 55439 | 6511 | 2637 | 2316 | 0.1382 |
| 0.3233 | 9.0 | 13113 | 0.4757 | 0.1790 | 0.1726 | 0.2612 | 0.7388 | 55414 | 6554 | 2619 | 2386 | 0.1401 |
| 0.3204 | 10.0 | 14570 | 0.4787 | 0.1786 | 0.1722 | 0.2608 | 0.7392 | 55434 | 6554 | 2599 | 2380 | 0.1399 |
Framework versions
- Transformers 4.21.2
- Pytorch 1.12.1+cu116
- Datasets 2.4.0
- Tokenizers 0.12.1
