<!-- 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-7front-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: 0.4563
- Wer: 0.1763
- Mer: 0.1699
- Wil: 0.2573
- Wip: 0.7427
- Hits: 55638
- Substitutions: 6453
- Deletions: 2496
- Insertions: 2439
- Cer: 0.1383
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6569 | 1.0 | 1457 | 0.4908 | 0.2308 | 0.2137 | 0.3042 | 0.6958 | 54844 | 6924 | 2819 | 5165 | 0.2114 |
| 0.5393 | 2.0 | 2914 | 0.4417 | 0.1916 | 0.1830 | 0.2706 | 0.7294 | 55263 | 6515 | 2809 | 3053 | 0.1559 |
| 0.4959 | 3.0 | 4371 | 0.4334 | 0.1776 | 0.1709 | 0.2574 | 0.7426 | 55660 | 6391 | 2536 | 2543 | 0.1398 |
| 0.4306 | 4.0 | 5828 | 0.4302 | 0.1755 | 0.1693 | 0.2563 | 0.7437 | 55618 | 6414 | 2555 | 2367 | 0.1372 |
| 0.3862 | 5.0 | 7285 | 0.4321 | 0.1752 | 0.1690 | 0.2561 | 0.7439 | 55653 | 6430 | 2504 | 2384 | 0.1369 |
| 0.3423 | 6.0 | 8742 | 0.4376 | 0.1772 | 0.1705 | 0.2576 | 0.7424 | 55660 | 6430 | 2497 | 2517 | 0.1384 |
| 0.3705 | 7.0 | 10199 | 0.4438 | 0.1746 | 0.1686 | 0.2559 | 0.7441 | 55589 | 6431 | 2567 | 2277 | 0.1358 |
| 0.2991 | 8.0 | 11656 | 0.4484 | 0.1763 | 0.1698 | 0.2571 | 0.7429 | 55650 | 6444 | 2493 | 2448 | 0.1382 |
| 0.2903 | 9.0 | 13113 | 0.4541 | 0.1764 | 0.1699 | 0.2570 | 0.7430 | 55655 | 6428 | 2504 | 2461 | 0.1384 |
| 0.2937 | 10.0 | 14570 | 0.4563 | 0.1763 | 0.1699 | 0.2573 | 0.7427 | 55638 | 6453 | 2496 | 2439 | 0.1383 |
Framework versions
- Transformers 4.21.2
- Pytorch 1.12.1+cu116
- Datasets 2.4.0
- Tokenizers 0.12.1
