wav2vec2-large-xls-r-300m-kr-jw4169

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the fleurs dataset. It achieves the following results on the evaluation set:

• Loss: 0.9752
• Wer: 0.5196

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: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
35.084	1.39	200	6.8536	1.0
4.853	2.78	400	4.6246	1.0
4.5491	4.17	600	4.3815	1.0
2.799	5.55	800	1.7402	0.8642
1.3872	6.94	1000	1.2019	0.7448
0.9599	8.33	1200	1.0594	0.7134
0.675	9.72	1400	0.9321	0.6404
0.4775	11.11	1600	0.9088	0.5911
0.3479	12.5	1800	0.9430	0.6010
0.2712	13.89	2000	0.8948	0.5854
0.2283	15.28	2200	0.9009	0.5495
0.1825	16.67	2400	0.9079	0.5501
0.161	18.06	2600	0.9518	0.5390
0.1394	19.44	2800	0.9529	0.5399
0.1266	20.83	3000	0.9505	0.5283
0.1102	22.22	3200	0.9748	0.5328
0.101	23.61	3400	0.9593	0.5316
0.0907	25.0	3600	0.9832	0.5292
0.0833	26.39	3800	0.9773	0.5181
0.0781	27.78	4000	0.9736	0.5163
0.0744	29.17	4200	0.9752	0.5196

Framework versions

• Transformers 4.25.0.dev0
• Pytorch 1.10.0+cu102
• Datasets 2.6.1
• Tokenizers 0.13.1