Hubert-kakeiken-W-closed_add_ver2

This model is a fine-tuned version of rinna/japanese-hubert-base on the ORIGINAL_KAKEIKEN_W_CLOSED_ADD_VER2 - JA dataset. It achieves the following results on the evaluation set:

• Loss: 0.0617
• Wer: 0.9988
• Cer: 1.0129

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
28.4059	1.0	880	10.6721	1.0	1.1284
9.1792	2.0	1760	6.9924	1.0	1.1284
4.9143	3.0	2640	3.8166	1.0	1.1284
3.1394	4.0	3520	2.8829	1.0	1.1283
2.7266	5.0	4400	1.9608	1.0	1.1444
1.4314	6.0	5280	0.8434	0.9999	1.0662
0.6837	7.0	6160	0.4583	0.9997	1.0330
0.403	8.0	7040	0.2512	0.9991	1.0479
0.3035	9.0	7920	0.1972	0.9993	1.0365
0.229	10.0	8800	0.0872	0.9991	1.0264
0.1995	11.0	9680	0.0959	0.9988	1.0262
0.1824	12.0	10560	0.1012	0.9988	1.0317
0.1774	13.0	11440	0.0541	0.9991	1.0220
0.1739	14.0	12320	0.0703	0.9990	1.0270
0.1609	15.0	13200	0.0480	0.9988	1.0203
0.1512	16.0	14080	0.0540	0.9988	1.0162
0.1412	17.0	14960	0.0396	0.9988	1.0188
0.1391	18.0	15840	0.0493	0.9988	1.0195
0.1325	19.0	16720	0.0366	0.9988	1.0186
0.1242	20.0	17600	0.0392	0.9988	1.0178
0.122	21.0	18480	0.0545	0.9988	1.0193
0.1143	22.0	19360	0.0408	0.9988	1.0185
0.1087	23.0	20240	0.0310	0.9988	1.0176
0.1013	24.0	21120	0.0262	0.9988	1.0166
0.0998	25.0	22000	0.0388	0.9988	1.0199
0.0903	26.0	22880	0.0280	0.9988	1.0166
0.088	27.0	23760	0.0492	0.9988	1.0197
0.0838	28.0	24640	0.0230	0.9988	1.0163
0.079	29.0	25520	0.0282	0.9988	1.0170
0.0747	30.0	26400	0.0271	0.9988	1.0162
0.0692	31.0	27280	0.0272	0.9988	1.0167
0.0699	32.0	28160	0.0427	0.9988	1.0143
0.0652	33.0	29040	0.0324	0.9988	1.0162
0.0624	34.0	29920	0.0315	0.9988	1.0163
0.0588	35.0	30800	0.0549	0.9988	1.0137
0.0594	36.0	31680	0.0457	0.9988	1.0142
0.0619	37.0	32560	0.0463	0.9988	1.0144
0.058	38.0	33440	0.0665	0.9988	1.0127
0.059	39.0	34320	0.0595	0.9988	1.0131
0.0563	39.9551	35160	0.0581	0.9988	1.0133

Framework versions

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