hubert-rinnna-jp-jdrtsp-fw07sp-clean

This model is a fine-tuned version of rinna/japanese-hubert-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2393
• Wer: 0.2187
• Cer: 0.1210

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: 2e-05
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
8.6058	1.0	164	6.2883	0.9743	0.9861
4.6616	2.0	328	4.0451	0.9743	0.9861
3.8526	3.0	492	3.5417	0.9743	0.9861
3.2384	4.0	656	3.0505	0.9743	0.9861
2.7948	5.0	820	2.6706	0.9743	0.9861
2.549	6.0	984	2.4268	0.9743	0.9861
2.1808	7.0	1148	1.8554	0.9743	0.9861
1.6069	8.0	1312	1.2551	0.6822	0.6231
1.1916	9.0	1476	0.7985	0.3679	0.2242
0.9977	10.0	1640	0.6234	0.3118	0.1827
0.836	11.0	1804	0.5103	0.2801	0.1643
0.7515	12.0	1968	0.4305	0.2663	0.1549
0.7045	13.0	2132	0.3688	0.2489	0.1413
0.6533	14.0	2296	0.3258	0.2399	0.1340
0.5906	15.0	2460	0.2941	0.2318	0.1288
0.5746	16.0	2624	0.2748	0.2300	0.1278
0.5169	17.0	2788	0.2573	0.2240	0.1242
0.5511	18.0	2952	0.2479	0.2211	0.1228
0.5318	19.0	3116	0.2410	0.2186	0.1210
0.5174	20.0	3280	0.2393	0.2187	0.1210

Framework versions

• Transformers 4.34.0.dev0
• Pytorch 2.0.1+cu118
• Datasets 2.14.5
• Tokenizers 0.13.3