wav2vec-arabic-V2-50

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: 0.3408
• Wer: 0.0460
• Per: 0.0347

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: 2
• eval_batch_size: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Per
10.1558	1.0	818	3.2108	1.0	1.0
3.0783	2.0	1636	2.0778	0.9260	0.9396
0.7602	3.0	2454	0.3836	0.1023	0.0870
0.2361	4.0	3272	0.3251	0.0668	0.0517
0.1602	5.0	4090	0.3240	0.0664	0.0515
0.1216	6.0	4908	0.3268	0.0673	0.0531
0.1078	7.0	5726	0.3501	0.0608	0.0465
0.0933	8.0	6544	0.3451	0.0538	0.0402
0.0713	9.0	7362	0.3658	0.0539	0.0407
0.0687	10.0	8180	0.3106	0.0519	0.0386
0.0561	11.0	8998	0.3322	0.0529	0.0396
0.0516	12.0	9816	0.3243	0.0484	0.0361
0.0392	13.0	10634	0.3412	0.0475	0.0354
0.037	14.0	11452	0.3370	0.0477	0.0359
0.0318	15.0	12270	0.3250	0.0466	0.0358
0.0291	16.0	13088	0.3451	0.0477	0.0359
0.025	17.0	13906	0.3713	0.0486	0.0368
0.0274	18.0	14724	0.3299	0.0459	0.0346
0.0208	19.0	15542	0.3451	0.0463	0.0349
0.0187	20.0	16360	0.3408	0.0460	0.0347

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0