hubert-rinnna-jp-jdrtsp-fw07sp-13

This model is a fine-tuned version of rinna/japanese-hubert-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1606
• Wer: 0.3004
• Cer: 0.1786

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: 0.0005
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.352	1.0	404	0.9913	0.6021	0.4479
0.9044	2.0	808	0.5053	0.4261	0.2774
0.9001	3.0	1212	0.8458	0.4848	0.3267
0.8425	4.0	1616	0.5311	0.4577	0.3053
0.8408	5.0	2020	0.4328	0.4075	0.2776
0.7759	6.0	2424	0.4736	0.4394	0.3363
0.7228	7.0	2828	0.4667	0.4173	0.2862
0.6755	8.0	3232	0.4190	0.4114	0.2611
0.634	9.0	3636	0.4252	0.3993	0.2612
0.6267	10.0	4040	0.3275	0.3734	0.2362
0.6199	11.0	4444	0.2786	0.3543	0.2222
0.5396	12.0	4848	0.2851	0.3501	0.2146
0.5343	13.0	5252	0.2527	0.3448	0.2106
0.5488	14.0	5656	0.2725	0.3431	0.2100
0.4606	15.0	6060	0.2293	0.3259	0.1962
0.4229	16.0	6464	0.2043	0.3172	0.1914
0.4078	17.0	6868	0.1891	0.3128	0.1862
0.4017	18.0	7272	0.1785	0.3075	0.1833
0.3618	19.0	7676	0.1673	0.3035	0.1803
0.3739	20.0	8080	0.1606	0.3004	0.1786

Framework versions

• Transformers 4.34.0.dev0
• Pytorch 2.0.1+cu118
• Datasets 2.14.5
• Tokenizers 0.13.3