<!-- 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-0front-1body-7rear
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: 0.4666
- Wer: 0.1780
- Mer: 0.1718
- Wil: 0.2607
- Wip: 0.7393
- Hits: 55410
- Substitutions: 6566
- Deletions: 2611
- Insertions: 2321
- Cer: 0.1388
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6424 | 1.0 | 1457 | 0.4944 | 0.1980 | 0.1893 | 0.2798 | 0.7202 | 54775 | 6748 | 3064 | 2975 | 0.1603 |
| 0.5444 | 2.0 | 2914 | 0.4496 | 0.1799 | 0.1740 | 0.2619 | 0.7381 | 55175 | 6480 | 2932 | 2207 | 0.1400 |
| 0.4975 | 3.0 | 4371 | 0.4451 | 0.1773 | 0.1713 | 0.2586 | 0.7414 | 55399 | 6429 | 2759 | 2266 | 0.1397 |
| 0.4312 | 4.0 | 5828 | 0.4417 | 0.1758 | 0.1701 | 0.2572 | 0.7428 | 55408 | 6407 | 2772 | 2178 | 0.1378 |
| 0.3846 | 5.0 | 7285 | 0.4445 | 0.1753 | 0.1696 | 0.2573 | 0.7427 | 55409 | 6453 | 2725 | 2142 | 0.1367 |
| 0.3501 | 6.0 | 8742 | 0.4482 | 0.1792 | 0.1727 | 0.2609 | 0.7391 | 55453 | 6522 | 2612 | 2439 | 0.1401 |
| 0.381 | 7.0 | 10199 | 0.4531 | 0.1770 | 0.1711 | 0.2592 | 0.7408 | 55380 | 6498 | 2709 | 2223 | 0.1378 |
| 0.313 | 8.0 | 11656 | 0.4585 | 0.1775 | 0.1716 | 0.2599 | 0.7401 | 55371 | 6516 | 2700 | 2250 | 0.1383 |
| 0.2976 | 9.0 | 13113 | 0.4646 | 0.1778 | 0.1717 | 0.2603 | 0.7397 | 55387 | 6537 | 2663 | 2284 | 0.1402 |
| 0.3152 | 10.0 | 14570 | 0.4666 | 0.1780 | 0.1718 | 0.2607 | 0.7393 | 55410 | 6566 | 2611 | 2321 | 0.1388 |
Framework versions
- Transformers 4.21.2
- Pytorch 1.12.1+cu116
- Datasets 2.4.0
- Tokenizers 0.12.1
