<!-- 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-1rear
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: 0.4600
- Wer: 0.1742
- Mer: 0.1683
- Wil: 0.2562
- Wip: 0.7438
- Hits: 55625
- Substitutions: 6495
- Deletions: 2467
- Insertions: 2291
- Cer: 0.1364
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6478 | 1.0 | 1457 | 0.4880 | 0.2256 | 0.2100 | 0.2999 | 0.7001 | 54825 | 6842 | 2920 | 4808 | 0.2019 |
| 0.542 | 2.0 | 2914 | 0.4461 | 0.1886 | 0.1807 | 0.2697 | 0.7303 | 55225 | 6615 | 2747 | 2817 | 0.1577 |
| 0.4873 | 3.0 | 4371 | 0.4390 | 0.1764 | 0.1702 | 0.2584 | 0.7416 | 55541 | 6519 | 2527 | 2344 | 0.1392 |
| 0.4271 | 4.0 | 5828 | 0.4361 | 0.1750 | 0.1691 | 0.2567 | 0.7433 | 55512 | 6453 | 2622 | 2226 | 0.1381 |
| 0.3705 | 5.0 | 7285 | 0.4366 | 0.1741 | 0.1684 | 0.2558 | 0.7442 | 55508 | 6427 | 2652 | 2164 | 0.1358 |
| 0.3557 | 6.0 | 8742 | 0.4424 | 0.1738 | 0.1679 | 0.2555 | 0.7445 | 55600 | 6453 | 2534 | 2235 | 0.1369 |
| 0.3838 | 7.0 | 10199 | 0.4471 | 0.1741 | 0.1684 | 0.2562 | 0.7438 | 55550 | 6473 | 2564 | 2210 | 0.1362 |
| 0.3095 | 8.0 | 11656 | 0.4517 | 0.1746 | 0.1685 | 0.2566 | 0.7434 | 55618 | 6499 | 2470 | 2305 | 0.1367 |
| 0.306 | 9.0 | 13113 | 0.4573 | 0.1748 | 0.1688 | 0.2570 | 0.7430 | 55601 | 6517 | 2469 | 2304 | 0.1369 |
| 0.3073 | 10.0 | 14570 | 0.4600 | 0.1742 | 0.1683 | 0.2562 | 0.7438 | 55625 | 6495 | 2467 | 2291 | 0.1364 |
Framework versions
- Transformers 4.21.2
- Pytorch 1.12.1+cu116
- Datasets 2.4.0
- Tokenizers 0.12.1
