metadata

license: cc-by-sa-4.0
tags:
  - generated_from_trainer
datasets:
  - te_dx_jp
model-index:
  - name: t5-base-TEDxJP-4front-1body-4rear
    results: []

t5-base-TEDxJP-4front-1body-4rear

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: 0.4398
Wer: 0.1697
Mer: 0.1641
Wil: 0.2506
Wip: 0.7494
Hits: 55824
Substitutions: 6360
Deletions: 2403
Insertions: 2197
Cer: 0.1335

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
0.5976	1.0	1457	0.4725	0.2267	0.2107	0.2985	0.7015	54877	6654	3056	4935	0.2024
0.4849	2.0	2914	0.4229	0.1789	0.1726	0.2590	0.7410	55401	6358	2828	2371	0.1432
0.4632	3.0	4371	0.4167	0.1725	0.1667	0.2529	0.7471	55723	6347	2517	2280	0.1343
0.3981	4.0	5828	0.4146	0.1716	0.1658	0.2521	0.7479	55784	6355	2448	2282	0.1336
0.3551	5.0	7285	0.4189	0.1713	0.1652	0.2512	0.7488	55909	6340	2338	2388	0.1345
0.3253	6.0	8742	0.4238	0.1714	0.1656	0.2514	0.7486	55805	6315	2467	2291	0.1359
0.308	7.0	10199	0.4292	0.1703	0.1645	0.2506	0.7494	55862	6341	2384	2271	0.1353
0.324	8.0	11656	0.4304	0.1693	0.1637	0.2497	0.7503	55856	6324	2407	2205	0.1336
0.2861	9.0	13113	0.4356	0.1694	0.1639	0.2501	0.7499	55814	6336	2437	2166	0.1332
0.2788	10.0	14570	0.4398	0.1697	0.1641	0.2506	0.7494	55824	6360	2403	2197	0.1335

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1