metadata

license: cc-by-sa-4.0
tags:
  - generated_from_trainer
datasets:
  - te_dx_jp
model-index:
  - name: t5-base-TEDxJP-0front-1body-10rear-order-RB
    results: []

t5-base-TEDxJP-0front-1body-10rear-order-RB

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: 0.4714
Wer: 0.1751
Mer: 0.1694
Wil: 0.2572
Wip: 0.7428
Hits: 55476
Substitutions: 6473
Deletions: 2638
Insertions: 2201
Cer: 0.1381

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: 30
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
0.6116	1.0	1457	0.4923	0.2289	0.2127	0.3015	0.6985	54722	6733	3132	4917	0.1992
0.5362	2.0	2914	0.4506	0.1835	0.1770	0.2661	0.7339	55105	6590	2892	2369	0.1447
0.4869	3.0	4371	0.4459	0.1806	0.1742	0.2629	0.7371	55298	6556	2733	2374	0.1424
0.4642	4.0	5828	0.4413	0.1767	0.1710	0.2588	0.7412	55331	6462	2794	2157	0.1379
0.4395	5.0	7285	0.4462	0.1779	0.1719	0.2594	0.7406	55367	6451	2769	2270	0.1391
0.3831	6.0	8742	0.4493	0.1751	0.1696	0.2568	0.7432	55370	6409	2808	2092	0.1369
0.3446	7.0	10199	0.4563	0.1769	0.1710	0.2595	0.7405	55401	6535	2651	2238	0.1397
0.3031	8.0	11656	0.4657	0.1754	0.1697	0.2578	0.7422	55436	6492	2659	2179	0.1372
0.3406	9.0	13113	0.4677	0.1750	0.1692	0.2570	0.7430	55502	6474	2611	2219	0.1365
0.3067	10.0	14570	0.4714	0.1751	0.1694	0.2572	0.7428	55476	6473	2638	2201	0.1381

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1