metadata

license: cc-by-sa-4.0
tags:
  - generated_from_trainer
datasets:
  - te_dx_jp
model-index:
  - name: t5-base-TEDxJP-8front-1body-8rear
    results: []

t5-base-TEDxJP-8front-1body-8rear

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: 0.4383
Wer: 0.1703
Mer: 0.1643
Wil: 0.2498
Wip: 0.7502
Hits: 55917
Substitutions: 6285
Deletions: 2385
Insertions: 2327
Cer: 0.1338

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
0.5634	1.0	1457	0.4668	0.2271	0.2101	0.2986	0.7014	55139	6756	2692	5219	0.1993
0.5335	2.0	2914	0.4168	0.1852	0.1776	0.2649	0.7351	55407	6467	2713	2782	0.1495
0.4453	3.0	4371	0.4124	0.1738	0.1678	0.2545	0.7455	55683	6391	2513	2321	0.1344
0.388	4.0	5828	0.4082	0.1703	0.1646	0.2502	0.7498	55838	6297	2452	2249	0.1324
0.3448	5.0	7285	0.4156	0.1704	0.1646	0.2505	0.7495	55840	6320	2427	2257	0.1339
0.3103	6.0	8742	0.4177	0.1690	0.1632	0.2484	0.7516	55955	6263	2369	2280	0.1324
0.3369	7.0	10199	0.4225	0.1688	0.1631	0.2480	0.7520	55930	6230	2427	2244	0.1327
0.3127	8.0	11656	0.4294	0.1692	0.1636	0.2489	0.7511	55876	6265	2446	2220	0.1331
0.2739	9.0	13113	0.4329	0.1702	0.1643	0.2501	0.7499	55903	6316	2368	2307	0.1338
0.269	10.0	14570	0.4383	0.1703	0.1643	0.2498	0.7502	55917	6285	2385	2327	0.1338

Framework versions

Transformers 4.21.2
Pytorch 1.12.1+cu116
Datasets 2.4.0
Tokenizers 0.12.1