fathyy/Graduation_Project_Distilation_Whisper1

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: 0.1259
• Wer: 0.2846

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
98.5205	1.0	313	0.1892	0.7493
29.6204	2.0	626	0.1465	0.4209
16.3912	3.0	939	0.1349	0.3646
10.2266	4.0	1252	0.1307	0.3240
7.221	5.0	1565	0.1326	0.3018
5.6073	6.0	1878	0.1318	0.2813
4.7685	7.0	2191	0.1276	0.2717
3.971	8.0	2504	0.1250	0.2701
3.5616	9.0	2817	0.1235	0.2698
3.2358	10.0	3130	0.1239	0.2753
3.0506	11.0	3443	0.1223	0.2738
2.8295	12.0	3756	0.1220	0.2782
2.5715	13.0	4069	0.1211	0.2722
2.4898	14.0	4382	0.1213	0.2683
2.4851	14.9536	4680	0.1213	0.2742

Framework versions

• Transformers 4.51.1
• Pytorch 2.5.1+cu124
• Datasets 3.5.0
• Tokenizers 0.21.0