Hubert-kakeiken-W-lounge

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

• Loss: 0.0115
• Wer: 0.9988
• Cer: 1.0143

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
40.3051	1.0	820	14.9753	1.0	1.1284
12.3062	2.0	1640	9.9534	1.0	1.1284
9.0113	3.0	2460	4.8377	1.0	1.1284
3.9478	4.0	3280	3.2016	1.0	1.1284
2.7396	5.0	4100	2.4831	1.0	1.1284
2.2509	6.0	4920	0.9198	0.9999	1.1054
0.6886	7.0	5740	0.3222	0.9991	1.0273
0.318	8.0	6560	0.1601	0.9991	1.0191
0.255	9.0	7380	0.0902	0.9988	1.0176
0.1835	10.0	8200	0.0716	0.9991	1.0197
0.16	11.0	9020	0.0492	0.9988	1.0175
0.1444	12.0	9840	0.0344	0.9990	1.0162
0.1307	13.0	10660	0.0388	0.9988	1.0176
0.1346	14.0	11480	0.0277	0.9990	1.0167
0.1257	15.0	12300	0.0267	0.9988	1.0164
0.1157	16.0	13120	0.0276	0.9990	1.0166
0.1104	17.0	13940	0.0341	0.9988	1.0172
0.1086	18.0	14760	0.0236	0.9988	1.0167
0.1072	19.0	15580	0.0306	0.9988	1.0175
0.128	20.0	16400	0.0241	0.9988	1.0160
0.0987	21.0	17220	0.0242	0.9988	1.0165
0.1031	22.0	18040	0.0229	0.9990	1.0159
0.0903	23.0	18860	0.0261	0.9988	1.0162
0.0895	24.0	19680	0.0322	0.9988	1.0177
0.0835	25.0	20500	0.0188	0.9988	1.0152
0.0744	26.0	21320	0.0179	0.9988	1.0149
0.0728	27.0	22140	0.0107	0.9988	1.0145
0.0704	28.0	22960	0.0161	0.9988	1.0149
0.068	29.0	23780	0.0140	0.9988	1.0150
0.0635	30.0	24600	0.0179	0.9988	1.0148
0.0606	31.0	25420	0.0170	0.9988	1.0146
0.0549	32.0	26240	0.0127	0.9988	1.0146
0.0557	33.0	27060	0.0112	0.9988	1.0146
0.0525	34.0	27880	0.0140	0.9988	1.0146
0.0478	35.0	28700	0.0125	0.9988	1.0146
0.0475	36.0	29520	0.0124	0.9988	1.0146
0.0455	37.0	30340	0.0114	0.9988	1.0142
0.0444	38.0	31160	0.0118	0.9988	1.0142
0.0476	39.0	31980	0.0119	0.9988	1.0142
0.0464	39.9518	32760	0.0115	0.9988	1.0143

Framework versions

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