Hubert-kakeiken-W-clean

This model is a fine-tuned version of rinna/japanese-hubert-base on the ORIGINAL_KAKEIKEN_W_CLEAN - JA dataset. It achieves the following results on the evaluation set:

• Loss: 0.0018
• Wer: 0.9990
• Cer: 1.0124

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: 3e-05
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 12500
• num_epochs: 40.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
20.8804	1.0	820	8.9051	1.0	1.1284
7.4869	2.0	1640	6.3826	1.0	1.1284
5.9007	3.0	2460	3.9252	1.0	1.1284
3.4438	4.0	3280	2.8383	1.0	1.1284
2.5168	5.0	4100	2.2606	1.0	1.1284
2.0175	6.0	4920	0.8359	1.0	1.0241
0.6013	7.0	5740	0.3640	0.9996	1.0556
0.2807	8.0	6560	0.1356	0.9991	1.0265
0.2171	9.0	7380	0.0684	0.9991	1.0165
0.1741	10.0	8200	0.0652	0.9991	1.0215
0.1395	11.0	9020	0.0215	0.9990	1.0140
0.135	12.0	9840	0.0659	0.9990	1.0144
0.1265	13.0	10660	0.0098	0.9991	1.0137
0.1244	14.0	11480	0.0161	0.9990	1.0142
0.1224	15.0	12300	0.0118	0.9990	1.0142
0.1143	16.0	13120	0.0063	0.9990	1.0132
0.1147	17.0	13940	0.0357	0.9988	1.0202
0.1168	18.0	14760	0.0061	0.9990	1.0132
0.0993	19.0	15580	0.0030	0.9990	1.0129
0.0982	20.0	16400	0.0092	0.9991	1.0131
0.0932	21.0	17220	0.0033	0.9990	1.0130
0.0878	22.0	18040	0.0063	0.9988	1.0134
0.0921	23.0	18860	0.0036	0.9990	1.0128
0.0802	24.0	19680	0.0022	0.9990	1.0127
0.0807	25.0	20500	0.0058	0.9993	1.0142
0.0757	26.0	21320	0.0103	0.9990	1.0142
0.071	27.0	22140	0.0040	0.9991	1.0128
0.0726	28.0	22960	0.0029	0.9990	1.0127
0.0625	29.0	23780	0.0053	0.9988	1.0135
0.0584	30.0	24600	0.0038	0.9988	1.0127
0.0612	31.0	25420	0.0029	0.9988	1.0130
0.057	32.0	26240	0.0047	0.9990	1.0128
0.0527	33.0	27060	0.0017	0.9990	1.0125
0.0491	34.0	27880	0.0017	0.9990	1.0124
0.0501	35.0	28700	0.0020	0.9990	1.0126
0.0458	36.0	29520	0.0019	0.9990	1.0124
0.0432	37.0	30340	0.0018	0.9990	1.0124
0.0451	38.0	31160	0.0018	0.9990	1.0124
0.0432	39.0	31980	0.0019	0.9991	1.0125
0.0453	39.9518	32760	0.0018	0.9991	1.0125

Framework versions

• Transformers 4.48.0
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.21.0