metadata

license: cc-by-sa-4.0
tags:
  - generated_from_trainer
datasets:
  - te_dx_jp
model-index:
  - name: t5-base-TEDxJP-10front-1body-10rear
    results: []

t5-base-TEDxJP-10front-1body-10rear

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: 0.4366
Wer: 0.1686
Mer: 0.1630
Wil: 0.2490
Wip: 0.7510
Hits: 55913
Substitutions: 6325
Deletions: 2349
Insertions: 2213
Cer: 0.1324

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
0.5904	1.0	1457	0.4553	0.2049	0.1941	0.2824	0.7176	54935	6595	3057	3580	0.1816
0.5001	2.0	2914	0.4201	0.1858	0.1776	0.2657	0.7343	55561	6554	2472	2973	0.1501
0.4615	3.0	4371	0.4099	0.1748	0.1685	0.2544	0.7456	55706	6326	2555	2410	0.1414
0.3988	4.0	5828	0.4040	0.1710	0.1654	0.2514	0.7486	55734	6319	2534	2189	0.1346
0.3859	5.0	7285	0.4131	0.1689	0.1635	0.2487	0.7513	55808	6245	2534	2129	0.1327
0.3259	6.0	8742	0.4138	0.1695	0.1639	0.2508	0.7492	55837	6400	2350	2198	0.1325
0.2915	7.0	10199	0.4233	0.1696	0.1637	0.2499	0.7501	55932	6344	2311	2297	0.1329
0.2638	8.0	11656	0.4298	0.1689	0.1633	0.2492	0.7508	55892	6319	2376	2213	0.1325
0.2888	9.0	13113	0.4321	0.1686	0.1630	0.2492	0.7508	55909	6343	2335	2210	0.1319
0.2614	10.0	14570	0.4366	0.1686	0.1630	0.2490	0.7510	55913	6325	2349	2213	0.1324

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1