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: 0.4411
Wer: 0.1694
Mer: 0.1636
Wil: 0.2489
Wip: 0.7511
Hits: 55916
Substitutions: 6268
Deletions: 2403
Insertions: 2267
Cer: 0.1371

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
0.5975	1.0	1457	0.4673	0.2058	0.1947	0.2831	0.7169	54961	6617	3009	3665	0.1746
0.5623	2.0	2914	0.4246	0.1801	0.1737	0.2603	0.7397	55335	6381	2871	2378	0.1452
0.4438	3.0	4371	0.4104	0.1735	0.1675	0.2541	0.7459	55705	6379	2503	2326	0.1357
0.4177	4.0	5828	0.4119	0.1704	0.1647	0.2504	0.7496	55837	6299	2451	2258	0.1338
0.3548	5.0	7285	0.4171	0.1690	0.1637	0.2487	0.7513	55785	6228	2574	2115	0.1328
0.3115	6.0	8742	0.4245	0.1687	0.1633	0.2484	0.7516	55838	6243	2506	2148	0.1340
0.2997	7.0	10199	0.4302	0.1706	0.1647	0.2503	0.7497	55898	6300	2389	2329	0.1351
0.2977	8.0	11656	0.4361	0.1699	0.1642	0.2496	0.7504	55879	6278	2430	2266	0.1342
0.2639	9.0	13113	0.4393	0.1688	0.1632	0.2486	0.7514	55891	6265	2431	2208	0.1365
0.2879	10.0	14570	0.4411	0.1694	0.1636	0.2489	0.7511	55916	6268	2403	2267	0.1371

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1