xls-r-300-cv17-polish-adap-cs

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4585
• Wer: 0.3182
• Cer: 0.0713

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.5986	1.6	100	3.9654	0.9986	0.9660
3.2886	3.2	200	3.4889	1.0	1.0
3.1683	4.8	300	3.1937	0.9946	0.9735
2.7362	6.4	400	2.6853	1.0	0.8424
0.6269	8.0	500	0.5183	0.5745	0.1381
0.2661	9.6	600	0.4218	0.4551	0.1048
0.1646	11.2	700	0.4160	0.4211	0.0985
0.1197	12.8	800	0.4793	0.4578	0.1072
0.1925	14.4	900	0.4402	0.4283	0.0969
0.1132	16.0	1000	0.4253	0.3909	0.0906
0.0851	17.6	1100	0.4609	0.3951	0.0921
0.0799	19.2	1200	0.4453	0.3944	0.0907
0.0657	20.8	1300	0.4681	0.3846	0.0887
0.1188	22.4	1400	0.4575	0.3785	0.0873
0.1088	24.0	1500	0.4649	0.3824	0.0882
0.0698	25.6	1600	0.4496	0.3611	0.0817
0.0575	27.2	1700	0.4459	0.3585	0.0822
0.0705	28.8	1800	0.4542	0.3608	0.0820
0.0524	30.4	1900	0.4785	0.3549	0.0814
0.0338	32.0	2000	0.4566	0.3521	0.0801
0.0357	33.6	2100	0.4597	0.3472	0.0783
0.0477	35.2	2200	0.4626	0.3451	0.0788
0.0478	36.8	2300	0.4730	0.3375	0.0765
0.0568	38.4	2400	0.4713	0.3333	0.0749
0.0217	40.0	2500	0.4701	0.3324	0.0755
0.0404	41.6	2600	0.4585	0.3278	0.0740
0.0118	43.2	2700	0.4656	0.3259	0.0736
0.0374	44.8	2800	0.4625	0.3249	0.0731
0.0417	46.4	2900	0.4599	0.3206	0.0721
0.0378	48.0	3000	0.4614	0.3195	0.0717
0.0381	49.6	3100	0.4585	0.3182	0.0713

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.3.1+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1