yesj1234/xlsr_enko_exp4

en-xlsr

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the ./SAMPLE_SPEECH.PY - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.5574
• Cer: 0.0835
• Wer: 0.1854

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: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.01
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
0.6992	2.79	600	0.4981	0.1370	0.3376
0.3394	5.58	1200	0.3934	0.1057	0.2467
0.2376	8.37	1800	0.4123	0.1015	0.2356
0.1877	11.16	2400	0.4269	0.0928	0.2136
0.1494	13.95	3000	0.4648	0.0922	0.2102
0.1186	16.74	3600	0.4835	0.0919	0.2058
0.0966	19.53	4200	0.4986	0.0875	0.1978
0.083	22.33	4800	0.5179	0.0862	0.1927
0.071	25.12	5400	0.5539	0.0857	0.1908
0.0648	27.91	6000	0.5583	0.0844	0.1870

Framework versions

• Transformers 4.34.1
• Pytorch 2.1.0+cu121
• Datasets 2.14.5
• Tokenizers 0.14.1