metadata

license: cc-by-sa-4.0
tags:
  - generated_from_trainer
datasets:
  - te_dx_jp
model-index:
  - name: t5-base-TEDxJP-7front-1body-7rear
    results: []

t5-base-TEDxJP-7front-1body-7rear

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: 0.4385
Wer: 0.1688
Mer: 0.1632
Wil: 0.2486
Wip: 0.7514
Hits: 55895
Substitutions: 6273
Deletions: 2419
Insertions: 2208
Cer: 0.1334

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: 10
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
0.5657	1.0	1457	0.4744	0.2212	0.2056	0.2946	0.7054	55195	6783	2609	4892	0.1922
0.5388	2.0	2914	0.4193	0.1791	0.1724	0.2601	0.7399	55526	6485	2576	2507	0.1404
0.447	3.0	4371	0.4119	0.1728	0.1670	0.2531	0.7469	55690	6334	2563	2266	0.1341
0.3952	4.0	5828	0.4088	0.1704	0.1647	0.2503	0.7497	55802	6283	2502	2220	0.1330
0.3498	5.0	7285	0.4138	0.1701	0.1643	0.2500	0.7500	55881	6305	2401	2280	0.1321
0.3107	6.0	8742	0.4201	0.1693	0.1636	0.2497	0.7503	55888	6334	2365	2236	0.1319
0.3443	7.0	10199	0.4239	0.1694	0.1637	0.2495	0.7505	55890	6309	2388	2241	0.1328
0.3099	8.0	11656	0.4316	0.1695	0.1639	0.2496	0.7504	55833	6292	2462	2192	0.1337
0.2804	9.0	13113	0.4345	0.1687	0.1631	0.2486	0.7514	55896	6273	2418	2206	0.1332
0.2714	10.0	14570	0.4385	0.1688	0.1632	0.2486	0.7514	55895	6273	2419	2208	0.1334

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1