<!-- 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-2rear
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: 0.4789
- Wer: 0.1778
- Mer: 0.1717
- Wil: 0.2602
- Wip: 0.7398
- Hits: 55410
- Substitutions: 6542
- Deletions: 2635
- Insertions: 2306
- Cer: 0.1406
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6615 | 1.0 | 1457 | 0.5044 | 0.2284 | 0.2125 | 0.3029 | 0.6971 | 54662 | 6871 | 3054 | 4826 | 0.1972 |
| 0.5561 | 2.0 | 2914 | 0.4658 | 0.1850 | 0.1783 | 0.2672 | 0.7328 | 55088 | 6583 | 2916 | 2451 | 0.1484 |
| 0.5114 | 3.0 | 4371 | 0.4555 | 0.1813 | 0.1748 | 0.2624 | 0.7376 | 55272 | 6463 | 2852 | 2392 | 0.1454 |
| 0.4419 | 4.0 | 5828 | 0.4542 | 0.1801 | 0.1737 | 0.2621 | 0.7379 | 55326 | 6529 | 2732 | 2372 | 0.1418 |
| 0.3992 | 5.0 | 7285 | 0.4541 | 0.1786 | 0.1725 | 0.2609 | 0.7391 | 55329 | 6527 | 2731 | 2276 | 0.1403 |
| 0.3649 | 6.0 | 8742 | 0.4598 | 0.1769 | 0.1710 | 0.2596 | 0.7404 | 55401 | 6540 | 2646 | 2239 | 0.1386 |
| 0.399 | 7.0 | 10199 | 0.4653 | 0.1786 | 0.1725 | 0.2610 | 0.7390 | 55327 | 6534 | 2726 | 2276 | 0.1414 |
| 0.3348 | 8.0 | 11656 | 0.4712 | 0.1784 | 0.1722 | 0.2611 | 0.7389 | 55391 | 6569 | 2627 | 2327 | 0.1409 |
| 0.3221 | 9.0 | 13113 | 0.4762 | 0.1782 | 0.1720 | 0.2608 | 0.7392 | 55404 | 6563 | 2620 | 2325 | 0.1408 |
| 0.3292 | 10.0 | 14570 | 0.4789 | 0.1778 | 0.1717 | 0.2602 | 0.7398 | 55410 | 6542 | 2635 | 2306 | 0.1406 |
Framework versions
- Transformers 4.21.2
- Pytorch 1.12.1+cu116
- Datasets 2.4.0
- Tokenizers 0.12.1
