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: 0.4376
• Wer: 0.1693
• Mer: 0.1635
• Wil: 0.2492
• Wip: 0.7508
• Hits: 55925
• Substitutions: 6304
• Deletions: 2358
• Insertions: 2270
• Cer: 0.1339

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: 40
• 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
0.6093	1.0	1457	0.4649	0.2573	0.2324	0.3203	0.6797	54884	6813	2890	6916	0.2380
0.5049	2.0	2914	0.4186	0.1791	0.1722	0.2595	0.7405	55594	6456	2537	2575	0.1432
0.4746	3.0	4371	0.4147	0.1741	0.1681	0.2539	0.7461	55665	6315	2607	2324	0.1392
0.4295	4.0	5828	0.4118	0.1723	0.1661	0.2523	0.7477	55884	6360	2343	2425	0.1342
0.3881	5.0	7285	0.4123	0.1696	0.1639	0.2496	0.7504	55896	6301	2390	2265	0.1397
0.3151	6.0	8742	0.4174	0.1687	0.1631	0.2482	0.7518	55924	6249	2414	2236	0.1329
0.2977	7.0	10199	0.4248	0.1674	0.1618	0.2466	0.7534	56006	6227	2354	2229	0.1321
0.2737	8.0	11656	0.4293	0.1685	0.1629	0.2485	0.7515	55898	6288	2401	2192	0.1340
0.2574	9.0	13113	0.4374	0.1683	0.1627	0.2480	0.7520	55930	6268	2389	2212	0.1329
0.2472	10.0	14570	0.4376	0.1693	0.1635	0.2492	0.7508	55925	6304	2358	2270	0.1339

Framework versions

• Transformers 4.21.2
• Pytorch 1.12.1+cu116
• Datasets 2.4.0
• Tokenizers 0.12.1