<!-- 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-8front-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: 0.4589
- Wer: 0.1739
- Mer: 0.1679
- Wil: 0.2545
- Wip: 0.7455
- Hits: 55667
- Substitutions: 6385
- Deletions: 2535
- Insertions: 2309
- Cer: 0.1363
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6586 | 1.0 | 1457 | 0.4812 | 0.2110 | 0.1994 | 0.2888 | 0.7112 | 54745 | 6712 | 3130 | 3789 | 0.1784 |
| 0.5246 | 2.0 | 2914 | 0.4383 | 0.1839 | 0.1770 | 0.2641 | 0.7359 | 55251 | 6428 | 2908 | 2544 | 0.1481 |
| 0.4795 | 3.0 | 4371 | 0.4327 | 0.1811 | 0.1740 | 0.2610 | 0.7390 | 55523 | 6438 | 2626 | 2631 | 0.1458 |
| 0.4224 | 4.0 | 5828 | 0.4328 | 0.1754 | 0.1693 | 0.2555 | 0.7445 | 55577 | 6338 | 2672 | 2318 | 0.1397 |
| 0.3755 | 5.0 | 7285 | 0.4351 | 0.1723 | 0.1668 | 0.2529 | 0.7471 | 55607 | 6326 | 2654 | 2150 | 0.1362 |
| 0.3538 | 6.0 | 8742 | 0.4413 | 0.1728 | 0.1670 | 0.2531 | 0.7469 | 55696 | 6341 | 2550 | 2271 | 0.1372 |
| 0.3686 | 7.0 | 10199 | 0.4455 | 0.1715 | 0.1659 | 0.2519 | 0.7481 | 55692 | 6319 | 2576 | 2180 | 0.1354 |
| 0.3004 | 8.0 | 11656 | 0.4518 | 0.1727 | 0.1668 | 0.2537 | 0.7463 | 55712 | 6400 | 2475 | 2281 | 0.1371 |
| 0.2914 | 9.0 | 13113 | 0.4564 | 0.1739 | 0.1678 | 0.2544 | 0.7456 | 55681 | 6378 | 2528 | 2323 | 0.1370 |
| 0.297 | 10.0 | 14570 | 0.4589 | 0.1739 | 0.1679 | 0.2545 | 0.7455 | 55667 | 6385 | 2535 | 2309 | 0.1363 |
Framework versions
- Transformers 4.21.2
- Pytorch 1.12.1+cu116
- Datasets 2.4.0
- Tokenizers 0.12.1
