Baselhany/test_basel

Whisper tiny AR - BH

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

• Loss: 0.1727
• Wer: 399.9312
• Cer: 210.6474

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: 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: 14
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
No log	0.6349	10	186.5657	9.8260	332.2476
No log	1.3175	20	186.5463	8.9427	332.5733
No log	1.9524	30	179.3050	7.4165	329.6417
No log	2.5714	40	173.6168	4.7760	313.5179
6.244	3.1905	50	173.0538	1.3689	295.1140
6.244	3.8254	60	119.1807	0.5697	184.8534
6.244	4.4444	70	113.7061	0.4699	172.4756
6.244	5.3175	80	126.2085	0.4148	175.2443
6.244	5.9524	90	127.4316	0.3749	230.9446
0.3668	6.5714	100	146.0299	0.3359	242.0195
0.3668	7.3175	110	217.5306	0.3022	297.0684
0.3668	7.9524	120	280.9163	0.2764	312.5407
0.3668	8.6349	130	232.8286	0.2581	364.9837
0.3668	9.3175	140	247.0200	0.2431	348.0456
0.2195	9.9524	150	282.1588	0.2310	503.4202
0.2195	10.6349	160	274.8204	0.2206	514.1694
0.2195	11.3175	170	289.0118	0.2122	520.3583
0.2195	11.9524	180	230.2466	0.2032	469.2182
0.2195	12.6349	190	193.6711	0.1966	419.8697
0.166	13.3175	200	0.1915	570.1954	277.9654
0.166	13.9524	210	0.1847	619.8697	296.8162

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0