wavlm-base-plus_zh_tw_ver2

This model is a fine-tuned version of microsoft/wavlm-base-plus on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 6.5278
• Wer: 1.0

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: 7.5e-05
• train_batch_size: 32
• eval_batch_size: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2000
• num_epochs: 100.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
82.628	2.5	500	79.5587	1.0
17.5635	5.0	1000	11.5929	1.0
6.4288	7.5	1500	6.4475	1.0
6.4092	10.0	2000	6.4579	1.0
6.3982	12.5	2500	6.4662	1.0
6.391	15.0	3000	6.4655	1.0
6.4097	17.5	3500	6.4691	1.0
6.3986	20.0	4000	6.4702	1.0
6.4069	22.5	4500	6.4761	1.0
6.4158	25.0	5000	6.4750	1.0
6.4117	27.5	5500	6.4816	1.0
6.4086	30.0	6000	6.4806	1.0
6.3992	32.5	6500	6.4872	1.0
6.3946	35.0	7000	6.4866	1.0
6.4212	37.5	7500	6.4895	1.0
6.4051	40.0	8000	6.4926	1.0
6.398	42.5	8500	6.5015	1.0
6.3967	45.0	9000	6.4960	1.0
6.4096	47.5	9500	6.5003	1.0
6.4068	50.0	10000	6.5026	1.0
6.4062	52.5	10500	6.5071	1.0
6.395	55.0	11000	6.5066	1.0
6.4079	57.5	11500	6.5093	1.0
6.411	60.0	12000	6.5106	1.0
6.4023	62.5	12500	6.5112	1.0
6.4053	65.0	13000	6.5143	1.0
6.4103	67.5	13500	6.5172	1.0
6.3899	70.0	14000	6.5182	1.0
6.4054	72.5	14500	6.5197	1.0
6.391	75.0	15000	6.5200	1.0
6.3988	77.5	15500	6.5220	1.0
6.4059	80.0	16000	6.5228	1.0
6.392	82.5	16500	6.5233	1.0
6.3947	85.0	17000	6.5253	1.0
6.3966	87.5	17500	6.5259	1.0
6.3905	90.0	18000	6.5264	1.0
6.4003	92.5	18500	6.5272	1.0
6.3877	95.0	19000	6.5275	1.0
6.3903	97.5	19500	6.5277	1.0
6.3944	100.0	20000	6.5278	1.0

Framework versions

• Transformers 4.28.0.dev0
• Pytorch 1.12.0+cu102
• Datasets 2.10.1
• Tokenizers 0.13.2