metadata

license: cc-by-sa-4.0
tags:
  - generated_from_trainer
datasets:
  - te_dx_jp
model-index:
  - name: t5-base-TEDxJP-5front-1body-5rear
    results: []

t5-base-TEDxJP-5front-1body-5rear

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: 0.4393
Wer: 0.1715
Mer: 0.1657
Wil: 0.2520
Wip: 0.7480
Hits: 55766
Substitutions: 6346
Deletions: 2475
Insertions: 2256
Cer: 0.1352

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
0.5978	1.0	1457	0.4748	0.2133	0.2003	0.2897	0.7103	55011	6755	2821	4200	0.1826
0.4837	2.0	2914	0.4190	0.1803	0.1737	0.2604	0.7396	55400	6393	2794	2459	0.1464
0.4576	3.0	4371	0.4114	0.1729	0.1673	0.2547	0.7453	55606	6443	2538	2187	0.1342
0.3888	4.0	5828	0.4169	0.1729	0.1671	0.2533	0.7467	55683	6347	2557	2264	0.1357
0.3647	5.0	7285	0.4178	0.1728	0.1666	0.2528	0.7472	55832	6358	2397	2403	0.1364
0.3226	6.0	8742	0.4199	0.1703	0.1648	0.2507	0.7493	55715	6304	2568	2125	0.1358
0.3045	7.0	10199	0.4309	0.1711	0.1653	0.2513	0.7487	55793	6322	2472	2257	0.1360
0.32	8.0	11656	0.4301	0.1714	0.1656	0.2518	0.7482	55768	6343	2476	2251	0.1341
0.2809	9.0	13113	0.4371	0.1706	0.1649	0.2513	0.7487	55774	6355	2458	2203	0.1333
0.2726	10.0	14570	0.4393	0.1715	0.1657	0.2520	0.7480	55766	6346	2475	2256	0.1352

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1