Baselhany/Distilation_Whisper_base_bigger_samples

Whisper base AR - BA

This model is a fine-tuned version of openai/whisper-base on the quran-ayat-speech-to-text dataset. It achieves the following results on the evaluation set:

• Loss: 0.0892
• Wer: 0.1918

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• 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
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.406	1.0	625	0.0887	0.1907
1.3322	2.0	1250	0.0906	0.1874
1.2587	3.0	1875	0.0903	0.1844
1.1135	4.0	2500	0.0892	0.1954
1.0444	5.0	3125	0.0879	0.1883
0.9344	6.0	3750	0.0867	0.1802
0.9135	7.0	4375	0.0874	0.1854
0.8567	8.0	5000	0.0861	0.1882
0.7738	9.0	5625	0.0857	0.1951
0.7419	10.0	6250	0.0852	0.1958
0.7167	11.0	6875	0.0854	0.1933
0.6929	12.0	7500	0.0850	0.1874
0.6539	13.0	8125	0.0847	0.1908
0.6448	14.0	8750	0.0845	0.1883
0.5887	15.0	9375	0.0846	0.1892

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1