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: 0.4713
Wer: 0.1763
Mer: 0.1704
Wil: 0.2586
Wip: 0.7414
Hits: 55456
Substitutions: 6510
Deletions: 2621
Insertions: 2256
Cer: 0.1383

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: 0
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
0.6725	1.0	1457	0.4909	0.2293	0.2133	0.3017	0.6983	54628	6686	3273	4851	0.2018
0.5083	2.0	2914	0.4537	0.1849	0.1781	0.2663	0.7337	55108	6513	2966	2464	0.1465
0.4943	3.0	4371	0.4466	0.1778	0.1716	0.2599	0.7401	55424	6519	2644	2319	0.1377
0.4454	4.0	5828	0.4385	0.1760	0.1703	0.2579	0.7421	55384	6452	2751	2163	0.1380
0.411	5.0	7285	0.4460	0.1755	0.1697	0.2570	0.7430	55466	6430	2691	2216	0.1379
0.3756	6.0	8742	0.4519	0.1750	0.1694	0.2568	0.7432	55419	6435	2733	2133	0.1383
0.3647	7.0	10199	0.4585	0.1755	0.1699	0.2579	0.7421	55368	6475	2744	2115	0.1379
0.3079	8.0	11656	0.4622	0.1763	0.1704	0.2590	0.7410	55416	6540	2631	2213	0.1387
0.3029	9.0	13113	0.4699	0.1762	0.1703	0.2584	0.7416	55451	6499	2637	2245	0.1386
0.2968	10.0	14570	0.4713	0.1763	0.1704	0.2586	0.7414	55456	6510	2621	2256	0.1383

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1