This model is a fine-tuned version of rinna/japanese-hubert-base on the MOZILLA-FOUNDATION/COMMON_VOICE_13_0 - JA dataset. It achieves the following results on the evaluation set:

Loss: 0.6997
Wer: 1.0
Cer: 0.3178

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: 16
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 32
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: 50.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	0.2660	100	43.2254	1.5295	5.8209
No log	0.5319	200	42.4473	1.5328	5.2847
No log	0.7979	300	40.4745	1.1351	1.8916
No log	1.0638	400	33.1315	1.0	0.9999
31.9649	1.3298	500	20.9238	1.0	0.9999
31.9649	1.5957	600	17.6015	1.0	0.9999
31.9649	1.8617	700	16.8689	1.0	0.9999
31.9649	2.1277	800	16.4459	1.0	0.9999
31.9649	2.3936	900	16.0021	1.0	0.9999
14.4801	2.6596	1000	15.5383	1.0	0.9999
14.4801	2.9255	1100	15.0419	1.0	0.9999
14.4801	3.1915	1200	14.5001	1.0	0.9999
14.4801	3.4574	1300	13.9273	1.0	0.9999
14.4801	3.7234	1400	13.3177	1.0	0.9999
12.1603	3.9894	1500	12.6793	1.0	0.9999
12.1603	4.2553	1600	12.0061	1.0	0.9999
12.1603	4.5213	1700	11.3136	1.0	0.9999
12.1603	4.7872	1800	10.5996	1.0	0.9999
12.1603	5.0532	1900	9.8780	1.0	0.9999
9.51	5.3191	2000	9.1535	1.0	0.9999
9.51	5.5851	2100	8.4424	1.0	0.9999
9.51	5.8511	2200	7.7545	1.0	0.9999
9.51	6.1170	2300	7.1014	1.0	0.9999
9.51	6.3830	2400	6.4994	1.0	0.9999
6.6851	6.6489	2500	5.9542	1.0	0.9999
6.6851	6.9149	2600	5.4843	1.0	0.9999
6.6851	7.1809	2700	5.0985	1.0	0.9999
6.6851	7.4468	2800	4.7872	1.0	0.9999
6.6851	7.7128	2900	4.5565	1.0	0.9999
4.7438	7.9787	3000	4.3885	1.0	0.9999
4.7438	8.2447	3100	4.2609	1.0	0.9999
4.7438	8.5106	3200	4.1909	1.0	0.9999
4.7438	8.7766	3300	4.1393	1.0	0.9999
4.7438	9.0426	3400	4.1096	1.0	0.9999
4.1247	9.3085	3500	4.0906	1.0	0.9999
4.1247	9.5745	3600	4.0835	1.0	0.9999
4.1247	9.8404	3700	4.0710	1.0	0.9999
4.1247	10.1064	3800	4.0627	1.0	0.9999
4.1247	10.3723	3900	4.0562	1.0	0.9999
4.039	10.6383	4000	4.0456	1.0	0.9999
4.039	10.9043	4100	4.0376	1.0	0.9999
4.039	11.1702	4200	4.0261	1.0	0.9999
4.039	11.4362	4300	4.0129	1.0	0.9999
4.039	11.7021	4400	4.0102	1.0	0.9999
3.9934	11.9681	4500	3.9765	1.0	0.9999
3.9934	12.2340	4600	3.9292	1.0	0.9999
3.9934	12.5	4700	3.8597	1.0	0.9999
3.9934	12.7660	4800	3.7626	1.0	0.9999
3.9934	13.0319	4900	3.6057	1.0	0.9999
3.7495	13.2979	5000	3.4160	1.0	0.9999
3.7495	13.5638	5100	3.1636	1.0	0.9998
3.7495	13.8298	5200	2.7153	1.0	0.7268
3.7495	14.0957	5300	2.4242	1.0	0.5731
3.7495	14.3617	5400	2.1946	1.0	0.5367
2.6537	14.6277	5500	2.0187	1.0	0.4742
2.6537	14.8936	5600	1.8818	1.0	0.4641
2.6537	15.1596	5700	1.7484	1.0	0.4355
2.6537	15.4255	5800	1.6397	1.0	0.4217
2.6537	15.6915	5900	1.5507	0.9998	0.4047
1.7492	15.9574	6000	1.4586	1.0	0.3999
1.7492	16.2234	6100	1.3908	1.0	0.3913
1.7492	16.4894	6200	1.3182	1.0	0.3849
1.7492	16.7553	6300	1.2754	1.0	0.3739
1.7492	17.0213	6400	1.2096	1.0	0.3744
1.3182	17.2872	6500	1.1599	1.0	0.3713
1.3182	17.5532	6600	1.1126	1.0	0.3673
1.3182	17.8191	6700	1.0685	1.0	0.3642
1.3182	18.0851	6800	1.0331	1.0	0.3606
1.3182	18.3511	6900	1.0006	1.0	0.3581
1.06	18.6170	7000	0.9663	1.0	0.3570
1.06	18.8830	7100	0.9409	1.0	0.3541
1.06	19.1489	7200	0.9141	0.9998	0.3527
1.06	19.4149	7300	0.8919	1.0	0.3515
1.06	19.6809	7400	0.8824	0.9998	0.3516
0.8957	19.9468	7500	0.8395	1.0	0.3477
0.8957	20.2128	7600	0.8356	1.0	0.3476
0.8957	20.4787	7700	0.8232	1.0	0.3506
0.8957	20.7447	7800	0.7880	1.0	0.3443
0.8957	21.0106	7900	0.7795	1.0	0.3453
0.7596	21.2766	8000	0.7670	1.0	0.3435
0.7596	21.5426	8100	0.7575	1.0	0.3426
0.7596	21.8085	8200	0.7344	1.0	0.3415
0.7596	22.0745	8300	0.7258	1.0	0.3413
0.7596	22.3404	8400	0.7230	0.9998	0.3410
0.6631	22.6064	8500	0.7223	1.0	0.3390
0.6631	22.8723	8600	0.7035	0.9998	0.3391
0.6631	23.1383	8700	0.7013	1.0	0.3370
0.6631	23.4043	8800	0.6926	1.0	0.3365
0.6631	23.6702	8900	0.6865	1.0	0.3368
0.6032	23.9362	9000	0.6834	1.0	0.3372
0.6032	24.2021	9100	0.6720	0.9998	0.3356
0.6032	24.4681	9200	0.6614	1.0	0.3353
0.6032	24.7340	9300	0.6691	1.0	0.3352
0.6032	25.0	9400	0.6571	0.9998	0.3328
0.544	25.2660	9500	0.6790	1.0	0.3360
0.544	25.5319	9600	0.6571	1.0	0.3326
0.544	25.7979	9700	0.6508	1.0	0.3344
0.544	26.0638	9800	0.6482	1.0	0.3327
0.544	26.3298	9900	0.6354	0.9998	0.3312
0.4943	26.5957	10000	0.6280	0.9998	0.3308
0.4943	26.8617	10100	0.6407	1.0	0.3299
0.4943	27.1277	10200	0.6388	1.0	0.3311
0.4943	27.3936	10300	0.6483	1.0	0.3302
0.4943	27.6596	10400	0.6255	0.9998	0.3311
0.4613	27.9255	10500	0.6355	1.0	0.3308
0.4613	28.1915	10600	0.6293	1.0	0.3300
0.4613	28.4574	10700	0.6333	1.0	0.3288
0.4613	28.7234	10800	0.6207	1.0	0.3279
0.4613	28.9894	10900	0.6218	0.9998	0.3281
0.4151	29.2553	11000	0.6301	1.0	0.3295
0.4151	29.5213	11100	0.6189	0.9996	0.3300
0.4151	29.7872	11200	0.6250	0.9998	0.3279
0.4151	30.0532	11300	0.6211	1.0	0.3268
0.4151	30.3191	11400	0.6195	1.0	0.3278
0.3895	30.5851	11500	0.6265	0.9998	0.3270
0.3895	30.8511	11600	0.6332	1.0	0.3263
0.3895	31.1170	11700	0.6232	0.9998	0.3253
0.3895	31.3830	11800	0.6281	1.0	0.3262
0.3895	31.6489	11900	0.6212	1.0	0.3258
0.3686	31.9149	12000	0.6368	0.9998	0.3266
0.3686	32.1809	12100	0.6276	0.9998	0.3299
0.3686	32.4468	12200	0.6513	1.0	0.3333
0.3686	32.7128	12300	0.6249	0.9996	0.3229
0.3686	32.9787	12400	0.6232	1.0	0.3262
0.3415	33.2447	12500	0.6144	1.0	0.3240
0.3415	33.5106	12600	0.6243	1.0	0.3269
0.3415	33.7766	12700	0.6344	1.0	0.3249
0.3415	34.0426	12800	0.6372	1.0	0.3236
0.3415	34.3085	12900	0.6399	1.0	0.3247
0.3167	34.5745	13000	0.6329	0.9996	0.3231
0.3167	34.8404	13100	0.6251	0.9998	0.3249
0.3167	35.1064	13200	0.6508	0.9998	0.3234
0.3167	35.3723	13300	0.6473	1.0	0.3219
0.3167	35.6383	13400	0.7159	0.9998	0.3232
0.3006	35.9043	13500	0.6520	0.9998	0.3279
0.3006	36.1702	13600	0.6568	0.9998	0.3225
0.3006	36.4362	13700	0.6568	1.0	0.3221
0.3006	36.7021	13800	0.6531	0.9998	0.3211
0.3006	36.9681	13900	0.6393	1.0	0.3218
0.2769	37.2340	14000	0.6671	0.9996	0.3229
0.2769	37.5	14100	0.6516	0.9998	0.3227
0.2769	37.7660	14200	0.6632	1.0	0.3218
0.2769	38.0319	14300	0.6799	1.0	0.3220
0.2769	38.2979	14400	0.6666	1.0	0.3220
0.2611	38.5638	14500	0.6486	1.0	0.3209
0.2611	38.8298	14600	0.6458	1.0	0.3209
0.2611	39.0957	14700	0.7051	1.0	0.3208
0.2611	39.3617	14800	0.6990	0.9998	0.3217
0.2611	39.6277	14900	0.6779	1.0	0.3216
0.2399	39.8936	15000	0.6609	1.0	0.3215
0.2399	40.1596	15100	0.6820	1.0	0.3208
0.2399	40.4255	15200	0.6618	1.0	0.3202
0.2399	40.6915	15300	0.6736	0.9998	0.3190
0.2399	40.9574	15400	0.6720	1.0	0.3196
0.2251	41.2234	15500	0.6697	1.0	0.3190
0.2251	41.4894	15600	0.6754	1.0	0.3214
0.2251	41.7553	15700	0.6777	1.0	0.3193
0.2251	42.0213	15800	0.6920	1.0	0.3195
0.2251	42.2872	15900	0.7009	1.0	0.3189
0.2181	42.5532	16000	0.6936	1.0	0.3184
0.2181	42.8191	16100	0.6941	1.0	0.3191
0.2181	43.0851	16200	0.6952	1.0	0.3190
0.2181	43.3511	16300	0.7078	1.0	0.3193
0.2181	43.6170	16400	0.6847	0.9998	0.3183
0.2019	43.8830	16500	0.6928	1.0	0.3203
0.2019	44.1489	16600	0.6941	1.0	0.3189
0.2019	44.4149	16700	0.6907	1.0	0.3180
0.2019	44.6809	16800	0.6850	0.9998	0.3183
0.2019	44.9468	16900	0.6956	1.0	0.3181
0.1938	45.2128	17000	0.6916	0.9998	0.3178
0.1938	45.4787	17100	0.6940	0.9996	0.3178
0.1938	45.7447	17200	0.6952	0.9998	0.3190
0.1938	46.0106	17300	0.7024	1.0	0.3189
0.1938	46.2766	17400	0.7070	0.9998	0.3188
0.1908	46.5426	17500	0.7046	0.9998	0.3182
0.1908	46.8085	17600	0.7046	0.9996	0.3183
0.1908	47.0745	17700	0.7030	1.0	0.3181
0.1908	47.3404	17800	0.6982	0.9998	0.3181
0.1908	47.6064	17900	0.7023	1.0	0.3177
0.1805	47.8723	18000	0.6954	1.0	0.3180
0.1805	48.1383	18100	0.6969	0.9998	0.3179
0.1805	48.4043	18200	0.6992	0.9998	0.3177
0.1805	48.6702	18300	0.6975	1.0	0.3176
0.1805	48.9362	18400	0.6991	0.9998	0.3177
0.1817	49.2021	18500	0.6993	0.9998	0.3175
0.1817	49.4681	18600	0.7004	0.9998	0.3177
0.1817	49.7340	18700	0.7000	1.0	0.3178
0.1817	50.0	18800	0.7000	0.9996	0.3176

Framework versions

Transformers 4.47.0.dev0
Pytorch 2.5.1+cu124
Datasets 3.1.0
Tokenizers 0.20.3

utakumi/Hubert-common_voice-ja-demo-kana-debug-50epochs-cosine

Hubert-common_voice-ja-demo-kana-debug-50epochs-cosine

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

utakumi/
Hubert-common_voice-ja-demo-kana-debug-50epochs-cosine