<!-- 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-6rear
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: 0.4688
- Wer: 0.1755
- Mer: 0.1695
- Wil: 0.2577
- Wip: 0.7423
- Hits: 55504
- Substitutions: 6505
- Deletions: 2578
- Insertions: 2249
- Cer: 0.1373
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6426 | 1.0 | 1457 | 0.4936 | 0.2128 | 0.2007 | 0.2903 | 0.7097 | 54742 | 6734 | 3111 | 3899 | 0.1791 |
| 0.5519 | 2.0 | 2914 | 0.4535 | 0.1970 | 0.1876 | 0.2747 | 0.7253 | 55096 | 6467 | 3024 | 3233 | 0.1567 |
| 0.5007 | 3.0 | 4371 | 0.4465 | 0.1819 | 0.1751 | 0.2628 | 0.7372 | 55359 | 6481 | 2747 | 2522 | 0.1435 |
| 0.4374 | 4.0 | 5828 | 0.4417 | 0.1761 | 0.1703 | 0.2582 | 0.7418 | 55399 | 6471 | 2717 | 2184 | 0.1373 |
| 0.3831 | 5.0 | 7285 | 0.4459 | 0.1755 | 0.1697 | 0.2570 | 0.7430 | 55465 | 6429 | 2693 | 2214 | 0.1383 |
| 0.352 | 6.0 | 8742 | 0.4496 | 0.1755 | 0.1697 | 0.2573 | 0.7427 | 55452 | 6450 | 2685 | 2202 | 0.1374 |
| 0.3955 | 7.0 | 10199 | 0.4527 | 0.1766 | 0.1707 | 0.2580 | 0.7420 | 55429 | 6429 | 2729 | 2251 | 0.1392 |
| 0.3132 | 8.0 | 11656 | 0.4629 | 0.1764 | 0.1703 | 0.2580 | 0.7420 | 55522 | 6472 | 2593 | 2329 | 0.1380 |
| 0.3116 | 9.0 | 13113 | 0.4652 | 0.1755 | 0.1695 | 0.2577 | 0.7423 | 55517 | 6505 | 2565 | 2264 | 0.1371 |
| 0.313 | 10.0 | 14570 | 0.4688 | 0.1755 | 0.1695 | 0.2577 | 0.7423 | 55504 | 6505 | 2578 | 2249 | 0.1373 |
Framework versions
- Transformers 4.21.2
- Pytorch 1.12.1+cu116
- Datasets 2.4.0
- Tokenizers 0.12.1
