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: 0.4373
Wer: 0.1699
Mer: 0.1642
Wil: 0.2499
Wip: 0.7501
Hits: 55848
Substitutions: 6297
Deletions: 2442
Insertions: 2236
Cer: 0.1360

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
0.5812	1.0	1457	0.4658	0.2393	0.2197	0.3076	0.6924	54882	6717	2988	5750	0.2187
0.5253	2.0	2914	0.4264	0.1832	0.1756	0.2632	0.7368	55549	6498	2540	2793	0.1520
0.4412	3.0	4371	0.4161	0.1728	0.1670	0.2535	0.7465	55665	6363	2559	2240	0.1360
0.3465	4.0	5828	0.4155	0.1706	0.1650	0.2504	0.7496	55756	6266	2565	2186	0.1356
0.3575	5.0	7285	0.4196	0.1696	0.1642	0.2498	0.7502	55781	6283	2523	2151	0.1358
0.3556	6.0	8742	0.4164	0.1687	0.1632	0.2487	0.7513	55857	6274	2456	2167	0.1341
0.3145	7.0	10199	0.4245	0.1705	0.1648	0.2504	0.7496	55819	6297	2471	2244	0.1355
0.3074	8.0	11656	0.4266	0.1693	0.1639	0.2494	0.7506	55799	6274	2514	2148	0.1358
0.269	9.0	13113	0.4352	0.1693	0.1637	0.2492	0.7508	55878	6288	2421	2225	0.1346
0.3162	10.0	14570	0.4373	0.1699	0.1642	0.2499	0.7501	55848	6297	2442	2236	0.1360

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1