wav2vec2-large-xls-r-300m-korean_y_freq

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

• Loss: 0.4522
• Cer: 13.5023

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: 0.0001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
35.2803	0.13	200	5.5453	100.0
4.6452	0.26	400	4.6065	100.0
4.5568	0.39	600	4.5732	97.9643
4.492	0.52	800	4.5091	97.9643
4.343	0.64	1000	4.2369	95.5012
3.3778	0.77	1200	2.7151	49.3917
2.4127	0.9	1400	2.2297	42.1509
2.0113	1.03	1600	1.8500	35.5926
1.7037	1.16	1800	1.5783	31.7942
1.502	1.29	2000	1.4251	29.2184
1.3376	1.42	2200	1.3812	30.4232
1.2586	1.55	2400	1.1952	25.6751
1.1415	1.68	2600	1.1614	25.8176
1.0858	1.81	2800	1.0985	24.9154
0.9846	1.93	3000	0.9985	21.8826
0.9443	2.06	3200	0.8759	19.9359
0.8298	2.19	3400	0.8549	19.5798
0.8111	2.32	3600	0.8271	22.2506
0.7507	2.45	3800	0.7808	20.9389
0.7202	2.58	4000	0.7409	19.5323
0.7121	2.71	4200	0.7139	18.5649
0.6639	2.84	4400	0.6726	17.4669
0.6595	2.97	4600	0.6830	16.6301
0.6161	3.09	4800	0.6334	15.5677
0.5604	3.22	5000	0.5943	14.4460
0.5387	3.35	5200	0.6169	15.4727
0.5317	3.48	5400	0.5831	15.0751
0.5171	3.61	5600	0.5414	14.3807
0.5	3.74	5800	0.5307	14.1789
0.4888	3.87	6000	0.5171	14.4875
0.4826	4.0	6200	0.5011	13.9712
0.4311	4.13	6400	0.5131	14.2679
0.4178	4.26	6600	0.4929	14.0364
0.3999	4.38	6800	0.4791	14.3451
0.4029	4.51	7000	0.4885	13.8584
0.3922	4.64	7200	0.4759	14.2323
0.3804	4.77	7400	0.4584	13.6091
0.3748	4.9	7600	0.4522	13.5023

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.0
• Tokenizers 0.15.0