MostafaKhidr/whisper-Large-ArabicEnglish

Whisper Large ArabicEnglish - Mostafa Khedr

This model is a fine-tuned version of openai/whisper-large-v2 on the arabic-english-code-switching dataset. It achieves the following results on the evaluation set:

• Loss: 0.7108
• Wer: 31.6184

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: 8
• eval_batch_size: 16
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH 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: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.558	0.6748	1000	0.5802	46.1378
0.3725	1.3495	2000	0.5258	43.1008
0.2131	2.0243	3000	0.5152	34.4890
0.204	2.6991	4000	0.5111	37.8727
0.1012	3.3738	5000	0.5475	34.1839
0.0593	4.0486	6000	0.5693	33.2686
0.0436	4.7233	7000	0.5895	33.0190
0.0189	5.3981	8000	0.6472	31.9235
0.0063	6.0729	9000	0.6850	32.2701
0.0046	6.7476	10000	0.7108	31.6184

Framework versions

• Transformers 4.46.3
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3