metadata

license: cc-by-sa-4.0
tags:
  - generated_from_trainer
datasets:
  - te_dx_jp
model-index:
  - name: t5-base-TEDxJP-0front-1body-6rear
    results: []

t5-base-TEDxJP-0front-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: 0.4698
Wer: 0.1756
Mer: 0.1696
Wil: 0.2578
Wip: 0.7422
Hits: 55507
Substitutions: 6509
Deletions: 2571
Insertions: 2259
Cer: 0.1373

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
0.6157	1.0	1457	0.4985	0.2141	0.2018	0.2933	0.7067	54712	6913	2962	3955	0.1920
0.5098	2.0	2914	0.4500	0.1847	0.1781	0.2663	0.7337	55040	6505	3042	2383	0.1473
0.4729	3.0	4371	0.4424	0.1768	0.1709	0.2591	0.7409	55397	6509	2681	2227	0.1386
0.4284	4.0	5828	0.4473	0.1753	0.1696	0.2573	0.7427	55457	6461	2669	2194	0.1362
0.394	5.0	7285	0.4462	0.1795	0.1729	0.2612	0.7388	55459	6530	2598	2468	0.1484
0.3556	6.0	8742	0.4521	0.1759	0.1700	0.2579	0.7421	55449	6480	2658	2221	0.1365
0.3348	7.0	10199	0.4588	0.1742	0.1686	0.2567	0.7433	55491	6491	2605	2156	0.1358
0.3594	8.0	11656	0.4608	0.1755	0.1697	0.2580	0.7420	55468	6516	2603	2215	0.1371
0.3065	9.0	13113	0.4685	0.1758	0.1698	0.2580	0.7420	55496	6510	2581	2263	0.1373
0.2976	10.0	14570	0.4698	0.1756	0.1696	0.2578	0.7422	55507	6509	2571	2259	0.1373

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1