Baselhany/Whisper_base_Quran_GP

Whisper base AR - BH

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: 0.0134
• Wer: 11.6358
• Cer: 3.5058

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: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 6
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
0.0124	0.2895	800	6.9720	0.0166	21.3510
0.0076	0.5790	1600	4.4857	0.0124	14.3371
0.0042	0.8685	2400	4.2342	0.0112	13.1816
0.0053	1.1581	3200	4.8224	0.0133	14.4143
0.0041	1.4476	4000	4.0206	0.0121	12.9768
0.0023	1.7371	4800	3.7118	0.0116	11.9643
0.0022	2.0268	5600	4.0467	0.0125	12.7101
0.002	2.3163	6400	3.7803	0.0125	12.1962
0.0016	2.6058	7200	3.7763	0.0124	12.2696
0.0018	2.8952	8000	3.6627	0.0122	12.0570
0.0013	3.1849	8800	3.6893	0.0126	12.0957
0.0015	3.4744	9600	3.6893	0.0126	12.2232
0.0013	3.7639	10400	3.6023	0.0124	11.8561
0.0009	4.0536	11200	3.6514	0.0127	11.9836
0.0009	4.3430	12000	3.5554	0.0125	11.6976
0.0008	4.6325	12800	3.4661	0.0130	11.5585
0.0009	4.9220	13600	3.4242	0.0130	11.4735
0.0007	5.2117	14400	0.0131	11.9102	3.5752
0.0008	5.5012	15200	0.0133	11.7595	3.5531
0.0008	5.7907	16000	0.0134	11.6358	3.5058

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.1
• Tokenizers 0.21.0