metadata

license: cc-by-sa-4.0
tags:
  - generated_from_trainer
datasets:
  - te_dx_jp
model-index:
  - name: t5-base-TEDxJP-7front-1body-7rear
    results: []

t5-base-TEDxJP-7front-1body-7rear

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: 0.4380
Wer: 0.1697
Mer: 0.1639
Wil: 0.2501
Wip: 0.7499
Hits: 55904
Substitutions: 6350
Deletions: 2333
Insertions: 2275
Cer: 0.1321

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
0.5926	1.0	1457	0.4717	0.2141	0.2008	0.2898	0.7102	55014	6714	2859	4253	0.1829
0.4821	2.0	2914	0.4178	0.1796	0.1733	0.2595	0.7405	55368	6348	2871	2384	0.1452
0.4444	3.0	4371	0.4103	0.1768	0.1700	0.2561	0.7439	55745	6359	2483	2577	0.1416
0.3824	4.0	5828	0.4145	0.1712	0.1653	0.2516	0.7484	55844	6362	2381	2314	0.1335
0.3481	5.0	7285	0.4133	0.1722	0.1659	0.2512	0.7488	55917	6283	2387	2449	0.1357
0.312	6.0	8742	0.4204	0.1719	0.1659	0.2516	0.7484	55845	6315	2427	2363	0.1360
0.3001	7.0	10199	0.4253	0.1684	0.1629	0.2486	0.7514	55908	6297	2382	2200	0.1312
0.3152	8.0	11656	0.4282	0.1689	0.1632	0.2491	0.7509	55909	6317	2361	2228	0.1322
0.2716	9.0	13113	0.4338	0.1694	0.1637	0.2497	0.7503	55865	6316	2406	2217	0.1321
0.2544	10.0	14570	0.4380	0.1697	0.1639	0.2501	0.7499	55904	6350	2333	2275	0.1321

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1