Hubert-kakeiken-W-closed_ver2

This model is a fine-tuned version of rinna/japanese-hubert-base on the ORIGINAL_KAKEIKEN_W_CLOSED_VER2 - JA dataset. It achieves the following results on the evaluation set:

• Loss: 0.0416
• Wer: 0.9988
• Cer: 1.0172

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
28.2151	1.0	644	11.5633	1.0	1.1284
11.2896	2.0	1288	9.1208	1.0	1.1284
9.1603	3.0	1932	6.1992	1.0	1.1283
4.9064	4.0	2576	3.9518	1.0	1.1284
3.6815	5.0	3220	3.1961	1.0	1.1283
3.1811	6.0	3864	2.5884	1.0	1.1284
2.1669	7.0	4508	1.7864	1.0	1.1439
1.4735	8.0	5152	0.8735	1.0	1.0649
0.9405	9.0	5796	0.5915	0.9999	1.0474
0.7037	10.0	6440	0.2996	0.9988	1.0013
0.3666	11.0	7084	0.1882	0.9988	1.0281
0.2964	12.0	7728	0.1888	0.9993	1.0349
0.2637	13.0	8372	0.1144	0.9988	1.0257
0.2265	14.0	9016	0.1531	0.9988	1.0305
0.213	15.0	9660	0.1029	0.9988	1.0232
0.1983	16.0	10304	0.0654	0.9990	1.0239
0.1941	17.0	10948	0.1301	0.9988	1.0309
0.1718	18.0	11592	0.0637	0.9988	1.0215
0.1774	19.0	12236	0.0641	0.9988	1.0226
0.1672	20.0	12880	0.0613	0.9988	1.0209
0.1509	21.0	13524	0.0480	0.9988	1.0212
0.1484	22.0	14168	0.0771	0.9988	1.0261
0.1474	23.0	14812	0.0673	0.9990	1.0225
0.1366	24.0	15456	0.0711	0.9988	1.0194
0.1334	25.0	16100	0.0962	0.9988	1.0187
0.1213	26.0	16744	0.0824	0.9988	1.0208
0.1166	27.0	17388	0.0571	0.9988	1.0198
0.1069	28.0	18032	0.1397	0.9988	1.0195
0.1022	29.0	18676	0.0408	0.9988	1.0184
0.0989	30.0	19320	0.0395	0.9988	1.0185
0.0925	31.0	19964	0.0451	0.9988	1.0184
0.0844	32.0	20608	0.0488	0.9988	1.0178
0.079	33.0	21252	0.0678	0.9988	1.0173
0.0801	34.0	21896	0.0525	0.9988	1.0170
0.0768	35.0	22540	0.0429	0.9988	1.0168
0.0661	36.0	23184	0.0500	0.9988	1.0173
0.0658	37.0	23828	0.0440	0.9988	1.0169
0.0659	38.0	24472	0.0437	0.9988	1.0171
0.0614	39.0	25116	0.0431	0.9988	1.0172
0.0613	39.9386	25720	0.0419	0.9988	1.0170

Framework versions

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