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: 0.4384
Wer: 0.1692
Mer: 0.1635
Wil: 0.2483
Wip: 0.7517
Hits: 55908
Substitutions: 6222
Deletions: 2457
Insertions: 2249
Cer: 0.1327

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
0.5881	1.0	1457	0.4572	0.2115	0.1995	0.2882	0.7118	54840	6661	3086	3916	0.1812
0.5018	2.0	2914	0.4167	0.1843	0.1765	0.2640	0.7360	55546	6494	2547	2863	0.1490
0.4633	3.0	4371	0.4110	0.1738	0.1679	0.2540	0.7460	55623	6327	2637	2260	0.1369
0.3971	4.0	5828	0.4068	0.1724	0.1666	0.2522	0.7478	55672	6278	2637	2218	0.1351
0.3907	5.0	7285	0.4131	0.1688	0.1635	0.2479	0.7521	55789	6180	2618	2106	0.1325
0.3305	6.0	8742	0.4147	0.1706	0.1649	0.2504	0.7496	55797	6281	2509	2227	0.1336
0.2937	7.0	10199	0.4236	0.1692	0.1636	0.2482	0.7518	55883	6207	2497	2223	0.1334
0.2649	8.0	11656	0.4307	0.1693	0.1638	0.2493	0.7507	55806	6272	2509	2154	0.1329
0.2914	9.0	13113	0.4319	0.1691	0.1634	0.2482	0.7518	55928	6230	2429	2262	0.1328
0.2598	10.0	14570	0.4384	0.1692	0.1635	0.2483	0.7517	55908	6222	2457	2249	0.1327

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1