arun100/whisper-small-ar-2

Whisper Small Arabic

This model is a fine-tuned version of arun100/whisper-small-ar-1 on the google/fleurs ar_eg dataset. It achieves the following results on the evaluation set:

• Loss: 0.4548
• Wer: 28.8090

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: 5e-07
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2414	52.0	500	0.3988	30.5694
0.0412	105.0	1000	0.4284	30.5694
0.0147	157.0	1500	0.4548	28.8090
0.0084	210.0	2000	0.4738	29.1125
0.0057	263.0	2500	0.4888	29.3553
0.0043	315.0	3000	0.5010	29.2218
0.0034	368.0	3500	0.5108	29.4889
0.0029	421.0	4000	0.5185	29.5010
0.0026	473.0	4500	0.5236	29.4889
0.0024	526.0	5000	0.5256	29.5375

Framework versions

• Transformers 4.38.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0