xlsr300m-korean-ctc

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.9963
• Wer: 0.5652
• Cer: 0.2894

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: 8
• eval_batch_size: 8
• seed: 1337
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.0396	0.9970	500	3.0211	1.0	0.9481
1.417	1.9930	1000	1.1934	0.8057	0.3569
1.2214	2.9890	1500	1.0721	0.6921	0.3182
1.2123	3.9850	2000	1.0590	0.6543	0.3205
1.1298	4.9811	2500	1.0206	0.6191	0.3044
1.1262	5.9771	3000	1.0117	0.6080	0.2961
1.0795	6.9731	3500	1.0168	0.5907	0.2934
1.0582	7.9691	4000	0.9936	0.5786	0.2965
1.0456	8.9651	4500	0.9955	0.5679	0.2905
1.0378	9.9611	5000	0.9963	0.5652	0.2894

Framework versions

• Transformers 4.56.2
• Pytorch 2.8.0+cu128
• Datasets 4.1.1
• Tokenizers 0.22.1