raghadOmar/whisper-quran

Whisper-raghadomar

This model is a fine-tuned version of tarteel-ai/whisper-base-ar-quran on the Zolfa Dataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.0325
• Wer: 10.3448

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: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0002	1.0	21	0.0317	10.3448
0.0002	2.0	42	0.0287	10.3448
0.0001	3.0	63	0.0293	10.3448
0.0002	4.0	84	0.0298	10.3448
0.0002	5.0	105	0.0281	10.3448
0.0002	6.0	126	0.0308	10.3448
0.0002	7.0	147	0.0262	10.3448
0.0008	8.0	168	0.0341	10.3448
0.0002	9.0	189	0.0223	3.4483
0.0003	10.0	210	0.0411	10.3448
0.0002	11.0	231	0.0357	10.3448
0.0003	12.0	252	0.0349	10.3448
0.0001	13.0	273	0.0429	10.3448
0.0003	14.0	294	0.0311	10.3448
0.0003	15.0	315	0.0372	10.3448
0.0002	16.0	336	0.0329	10.3448
0.0002	17.0	357	0.0390	10.3448
0.0004	18.0	378	0.0333	10.3448
0.0002	19.0	399	0.0450	10.3448
0.0003	20.0	420	0.0384	10.3448
0.0002	21.0	441	0.0366	10.3448
0.0002	22.0	462	0.0360	10.3448
0.0001	23.0	483	0.0441	10.3448
0.0006	23.8095	500	0.0325	10.3448

Framework versions

• Transformers 4.40.2
• Pytorch 2.2.1+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1