metadata

license: cc-by-sa-4.0
tags:
  - generated_from_trainer
datasets:
  - te_dx_jp
model-index:
  - name: t5-base-TEDxJP-5front-1body-5rear
    results: []

t5-base-TEDxJP-5front-1body-5rear

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: 0.4392
Wer: 0.1702
Mer: 0.1644
Wil: 0.2508
Wip: 0.7492
Hits: 55879
Substitutions: 6371
Deletions: 2337
Insertions: 2282
Cer: 0.1330

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
0.5939	1.0	1457	0.4718	0.2101	0.1989	0.2858	0.7142	54640	6461	3486	3620	0.1750
0.5477	2.0	2914	0.4248	0.1821	0.1750	0.2625	0.7375	55437	6468	2682	2613	0.1517
0.4481	3.0	4371	0.4129	0.1726	0.1668	0.2544	0.7456	55654	6453	2480	2212	0.1350
0.3919	4.0	5828	0.4108	0.1719	0.1661	0.2526	0.7474	55754	6367	2466	2271	0.1349
0.353	5.0	7285	0.4125	0.1693	0.1637	0.2493	0.7507	55856	6287	2444	2206	0.1336
0.3367	6.0	8742	0.4194	0.1696	0.1639	0.2493	0.7507	55886	6271	2430	2256	0.1324
0.2959	7.0	10199	0.4274	0.1698	0.1643	0.2502	0.7498	55810	6318	2459	2193	0.1343
0.2979	8.0	11656	0.4304	0.1711	0.1652	0.2516	0.7484	55871	6367	2349	2338	0.1337
0.2714	9.0	13113	0.4363	0.1694	0.1637	0.2501	0.7499	55884	6354	2349	2239	0.1319
0.2797	10.0	14570	0.4392	0.1702	0.1644	0.2508	0.7492	55879	6371	2337	2282	0.1330

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1