wav2vec-arabic-V2-5

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: 1.2918
• Wer: 0.4135
• Per: 0.3941

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
25.2164	1.0	82	9.3224	1.0	1.0
5.4912	2.0	164	3.4201	1.0	1.0
3.2902	3.0	246	3.2384	1.0	1.0
3.2447	4.0	328	3.2071	1.0	1.0
3.2259	5.0	410	3.2246	1.0	1.0
3.2254	6.0	492	3.2056	1.0	1.0
3.2116	7.0	574	3.2068	1.0	1.0
3.1886	8.0	656	3.2814	1.0	1.0
3.1499	9.0	738	3.1154	1.0	1.0
3.0862	10.0	820	3.0623	1.0	1.0
2.9718	11.0	902	2.9404	0.9973	0.9983
2.81	12.0	984	2.7654	0.9821	0.9859
2.612	13.0	1066	2.6608	0.9751	0.9785
2.4167	14.0	1148	2.2973	0.9051	0.9153
2.1839	15.0	1230	2.0349	0.7676	0.7781
1.8957	16.0	1312	1.7675	0.6932	0.6950
1.6773	17.0	1394	1.5461	0.5751	0.5707
1.5059	18.0	1476	1.4200	0.4381	0.4165
1.4219	19.0	1558	1.3218	0.4256	0.4049
1.3273	20.0	1640	1.2918	0.4135	0.3941

Framework versions

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