metadata

license: cc-by-sa-4.0
tags:
  - generated_from_trainer
datasets:
  - te_dx_jp
model-index:
  - name: t5-base-TEDxJP-4front-1body-4rear
    results: []

t5-base-TEDxJP-4front-1body-4rear

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: 0.4413
Wer: 0.1701
Mer: 0.1644
Wil: 0.2507
Wip: 0.7493
Hits: 55851
Substitutions: 6351
Deletions: 2385
Insertions: 2250
Cer: 0.1341

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
0.6042	1.0	1457	0.4733	0.2042	0.1938	0.2829	0.7171	54861	6662	3064	3461	0.1725
0.5747	2.0	2914	0.4297	0.1812	0.1744	0.2610	0.7390	55403	6393	2791	2517	0.1458
0.4509	3.0	4371	0.4187	0.1751	0.1686	0.2557	0.7443	55743	6433	2411	2464	0.1393
0.4216	4.0	5828	0.4133	0.1715	0.1656	0.2515	0.7485	55811	6324	2452	2301	0.1346
0.356	5.0	7285	0.4219	0.1703	0.1649	0.2502	0.7498	55717	6262	2608	2129	0.1327
0.3174	6.0	8742	0.4232	0.1700	0.1642	0.2494	0.7506	55863	6253	2471	2253	0.1331
0.2992	7.0	10199	0.4309	0.1701	0.1647	0.2511	0.7489	55746	6350	2491	2148	0.1332
0.307	8.0	11656	0.4317	0.1699	0.1644	0.2505	0.7495	55801	6331	2455	2189	0.1336
0.2692	9.0	13113	0.4385	0.1699	0.1643	0.2504	0.7496	55834	6334	2419	2223	0.1346
0.2965	10.0	14570	0.4413	0.1701	0.1644	0.2507	0.7493	55851	6351	2385	2250	0.1341

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1