metadata

license: cc-by-sa-4.0
tags:
  - generated_from_trainer
datasets:
  - te_dx_jp
model-index:
  - name: t5-base-TEDxJP-6front-1body-6rear
    results: []

t5-base-TEDxJP-6front-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: 0.4394
Wer: 0.1715
Mer: 0.1655
Wil: 0.2514
Wip: 0.7486
Hits: 55840
Substitutions: 6324
Deletions: 2423
Insertions: 2327
Cer: 0.1370

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
0.5934	1.0	1457	0.4618	0.2350	0.2167	0.3040	0.6960	54861	6643	3083	5449	0.2143
0.5143	2.0	2914	0.4200	0.1809	0.1738	0.2621	0.7379	55519	6548	2520	2613	0.1457
0.4671	3.0	4371	0.4138	0.1726	0.1669	0.2535	0.7465	55651	6368	2568	2212	0.1349
0.4044	4.0	5828	0.4077	0.1708	0.1653	0.2518	0.7482	55708	6359	2520	2155	0.1371
0.398	5.0	7285	0.4140	0.1691	0.1638	0.2496	0.7504	55749	6294	2544	2083	0.1329
0.3394	6.0	8742	0.4190	0.1701	0.1645	0.2511	0.7489	55822	6375	2390	2224	0.1348
0.3009	7.0	10199	0.4264	0.1720	0.1659	0.2519	0.7481	55861	6341	2385	2381	0.1369
0.273	8.0	11656	0.4307	0.1703	0.1647	0.2509	0.7491	55772	6337	2478	2184	0.1352
0.2939	9.0	13113	0.4350	0.1697	0.1640	0.2499	0.7501	55843	6313	2431	2214	0.1353
0.2747	10.0	14570	0.4394	0.1715	0.1655	0.2514	0.7486	55840	6324	2423	2327	0.1370

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1