sarh21/whisper-small-ar

Whisper Small Arabic - Fine-tuned on Omnilingual ASR Corpus

This model is a fine-tuned version of openai/whisper-small on the Omnilingual ASR Corpus dataset. It achieves the following results on the evaluation set:

• Loss: 1.3057
• Wer: 50.5491

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3011	6.2893	1000	0.8454	49.5801
0.0419	12.5786	2000	1.0918	49.4832
0.0055	18.8679	3000	1.2586	52.7939
0.0026	25.1572	4000	1.3057	50.5491

Framework versions

• Transformers 5.1.0
• Pytorch 2.9.0+cu126
• Datasets 3.3.2
• Tokenizers 0.22.2