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: inf
Wer: 1.0
Cer: 0.2359

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: 20.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	0.2660	100	inf	1.8204	4.9067
No log	0.5319	200	inf	1.5926	4.6323
No log	0.7979	300	inf	1.1770	1.9637
No log	1.0638	400	inf	1.0	0.9817
14.493	1.3298	500	inf	1.0	0.9817
14.493	1.5957	600	inf	1.0	0.9817
14.493	1.8617	700	inf	1.0	0.9817
14.493	2.1277	800	inf	1.0	0.9817
14.493	2.3936	900	inf	1.0	0.9817
6.5744	2.6596	1000	6.8080	1.0	0.9817
6.5744	2.9255	1100	6.5972	1.0	0.9817
6.5744	3.1915	1200	inf	1.0	0.9817
6.5744	3.4574	1300	inf	1.0	0.9817
6.5744	3.7234	1400	inf	1.0	0.9817
5.5193	3.9894	1500	inf	1.0	0.9817
5.5193	4.2553	1600	inf	1.0	0.9817
5.5193	4.5213	1700	inf	1.0	0.9817
5.5193	4.7872	1800	inf	1.0	0.9817
5.5193	5.0532	1900	inf	1.0	0.9817
4.5578	5.3191	2000	inf	1.0	0.9817
4.5578	5.5851	2100	inf	1.0	0.9817
4.5578	5.8511	2200	inf	1.0	0.9817
4.5578	6.1170	2300	inf	1.0	0.9817
4.5578	6.3830	2400	inf	1.0	0.9817
3.6943	6.6489	2500	inf	1.0	0.9817
3.6943	6.9149	2600	inf	1.0	0.9817
3.6943	7.1809	2700	inf	1.0	0.9817
3.6943	7.4468	2800	inf	1.0	0.9817
3.6943	7.7128	2900	3.1572	1.0	0.9817
3.1932	7.9787	3000	inf	1.0	0.9817
3.1932	8.2447	3100	inf	1.0	0.9817
3.1932	8.5106	3200	inf	1.0	0.9817
3.1932	8.7766	3300	inf	1.0	0.9817
3.1932	9.0426	3400	inf	1.0	0.9817
3.0309	9.3085	3500	inf	1.0	0.9817
3.0309	9.5745	3600	inf	1.0	0.9817
3.0309	9.8404	3700	inf	1.0	0.9817
3.0309	10.1064	3800	inf	1.0	0.9817
3.0309	10.3723	3900	inf	1.0	0.9817
2.9704	10.6383	4000	inf	1.0	0.9817
2.9704	10.9043	4100	inf	1.0	0.9817
2.9704	11.1702	4200	inf	1.0	0.9049
2.9704	11.4362	4300	inf	1.0	0.7254
2.9704	11.7021	4400	inf	1.0	0.4365
2.2767	11.9681	4500	1.5675	1.0	0.3732
2.2767	12.2340	4600	inf	1.0	0.3455
2.2767	12.5	4700	inf	1.0	0.3277
2.2767	12.7660	4800	inf	1.0	0.3053
2.2767	13.0319	4900	inf	1.0	0.2935
1.2873	13.2979	5000	inf	1.0	0.2784
1.2873	13.5638	5100	inf	1.0	0.2684
1.2873	13.8298	5200	inf	1.0	0.2678
1.2873	14.0957	5300	inf	1.0	0.2616
1.2873	14.3617	5400	0.8214	1.0	0.2608
0.9318	14.6277	5500	inf	1.0	0.2564
0.9318	14.8936	5600	inf	1.0	0.2544
0.9318	15.1596	5700	inf	1.0	0.2525
0.9318	15.4255	5800	inf	1.0	0.2510
0.9318	15.6915	5900	inf	1.0	0.2527
0.754	15.9574	6000	inf	1.0	0.2499
0.754	16.2234	6100	0.6672	1.0	0.2485
0.754	16.4894	6200	inf	1.0	0.2464
0.754	16.7553	6300	inf	1.0	0.2467
0.754	17.0213	6400	inf	1.0	0.2411
0.6421	17.2872	6500	inf	1.0	0.2411
0.6421	17.5532	6600	inf	1.0	0.2418
0.6421	17.8191	6700	inf	1.0	0.2386
0.6421	18.0851	6800	inf	0.9996	0.2387
0.6421	18.3511	6900	inf	1.0	0.2381
0.568	18.6170	7000	inf	1.0	0.2391
0.568	18.8830	7100	inf	1.0	0.2370
0.568	19.1489	7200	inf	1.0	0.2344
0.568	19.4149	7300	inf	1.0	0.2364
0.568	19.6809	7400	inf	1.0	0.2347
0.5259	19.9468	7500	inf	1.0	0.2334

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-phonemes-cosine-3e-5

Hubert-common_voice-ja-demo-phonemes-cosine-3e-5

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

utakumi/
Hubert-common_voice-ja-demo-phonemes-cosine-3e-5