mark-muhammad/tarteel-base-finetuned

Tarteel Base with Makharij Huruf

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

• Loss: 0.0206
• Wer: 0.0187

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: 3e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 50
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
42.9673	1.0	20	2.8351	0.8667
49.3440	2.0	40	3.4646	0.8780
33.1581	3.0	60	2.1838	0.8755
28.3724	4.0	80	1.7310	0.8767
13.9161	5.0	100	0.8567	0.8369
7.7049	6.0	120	0.4305	0.8742
4.7890	7.0	140	0.2922	0.3400
4.4606	8.0	160	0.2770	0.2939
2.9499	9.0	180	0.1666	0.9054
2.8042	10.0	200	0.1659	0.2279
2.0015	11.0	220	0.1121	0.1544
2.0761	12.0	240	0.1155	0.1557
0.9960	13.0	260	0.0682	0.0785
0.7975	14.0	280	0.0573	0.0872
0.8083	15.0	300	0.0473	0.0797
0.7392	16.0	320	0.0520	0.0897
0.7226	17.0	340	0.0450	0.0610
0.5558	18.0	360	0.0413	0.0598
0.3268	19.0	380	0.0342	0.0560
0.3140	20.0	400	0.0303	0.0548
0.3304	21.0	420	0.0296	0.0523
0.3635	22.0	440	0.0298	0.0361
0.3152	23.0	460	0.0276	0.0324
0.4463	24.0	480	0.0267	0.0336
0.3206	25.0	500	0.0231	0.0286
0.4211	26.0	520	0.0206	0.0187
0.2552	27.0	540	0.0201	0.0187
0.2434	28.0	560	0.0201	0.0187
0.1776	29.0	580	0.0201	0.0187
0.4475	30.0	600	0.0201	0.0187

Framework versions

• Transformers 5.8.1
• Pytorch 2.12.0
• Datasets 4.8.5
• Tokenizers 0.22.2