wav2vec2-large-xlsr-korean-demo-colab_epoch15

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4133
• Wer: 0.3801

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: 0.0003
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
16.9017	0.8	400	4.6273	1.0
4.4633	1.6	800	4.4419	1.0
4.2262	2.4	1200	3.8477	0.9994
2.4402	3.21	1600	1.3564	0.8111
1.3499	4.01	2000	0.9070	0.6664
0.9922	4.81	2400	0.7496	0.6131
0.8271	5.61	2800	0.6240	0.5408
0.6918	6.41	3200	0.5506	0.5026
0.6015	7.21	3600	0.5303	0.4935
0.5435	8.02	4000	0.4951	0.4696
0.4584	8.82	4400	0.4677	0.4432
0.4258	9.62	4800	0.4602	0.4307
0.3906	10.42	5200	0.4456	0.4195
0.3481	11.22	5600	0.4265	0.4062
0.3216	12.02	6000	0.4241	0.4046
0.2908	12.83	6400	0.4106	0.3941
0.2747	13.63	6800	0.4146	0.3855
0.2633	14.43	7200	0.4133	0.3801

Framework versions

• Transformers 4.17.0
• Pytorch 1.12.0+cu113
• Datasets 1.18.3
• Tokenizers 0.12.1