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: 0.0847
Wer: 0.1936

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: 30
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.1291	0.5858	1000	0.0912	0.1978
1.7057	1.1716	2000	0.0912	0.2003
1.7162	1.7575	3000	0.0912	0.2060
1.4996	2.3433	4000	0.0901	0.2047
1.3942	2.9291	5000	0.0883	0.1951
1.2285	3.5149	6000	0.0876	0.1957
1.0637	4.1008	7000	0.0873	0.1920
1.1144	4.6866	8000	0.0865	0.1927
1.0164	5.2724	9000	0.0858	0.1923
0.9812	5.8582	10000	0.0856	0.1941
0.8927	6.4441	11000	0.0849	0.2017
0.8936	7.0299	12000	0.0844	0.1961
0.8718	7.6157	13000	0.0854	0.1979
0.9019	8.2015	14000	0.0847	0.1854
0.8293	8.7873	15000	0.0847	0.1983
0.8363	9.3732	16000	0.0842	0.1982
0.8034	9.9590	17000	0.0840	0.1975
0.8462	10.5448	18000	0.0855	0.1953
0.8824	11.1306	19000	0.0848	0.1930
0.8591	11.7165	20000	0.0849	0.1838
0.8339	12.3023	21000	0.0842	0.1863
0.8573	12.8881	22000	0.0836	0.1926
0.7445	13.4739	23000	0.0839	0.1842
0.783	14.0598	24000	0.0836	0.1842
0.7263	14.6456	25000	0.0839	0.1824
0.7634	15.2314	26000	0.0835	0.1826
0.7379	15.8172	27000	0.0834	0.1829
0.7902	16.4030	28000	0.0842	0.1811
0.8261	16.9889	29000	0.0841	0.1849
0.7531	17.5747	30000	0.0840	0.1867
0.7166	18.1605	31000	0.0839	0.1905
0.7976	18.7463	32000	0.0841	0.1838
0.7008	19.3322	33000	0.0835	0.1864
0.707	19.9180	34000	0.0833	0.1872
0.6865	20.5038	35000	0.0835	0.1844
0.6927	21.0896	36000	0.0834	0.1882
0.7014	21.6755	37000	0.0835	0.1861
0.6951	22.2613	38000	0.0833	0.1874
0.6848	22.8471	39000	0.0834	0.1927
0.7096	23.4329	40000	0.0834	0.1936
0.6952	24.0187	41000	0.0835	0.1933
0.692	24.6046	42000	0.0833	0.1930
0.6552	25.1904	43000	0.0831	0.1867
0.6641	25.7762	44000	0.0832	0.1874
0.6921	26.3620	45000	0.0833	0.1880
0.6894	26.9479	46000	0.0832	0.1855
0.7041	27.5337	47000	0.0827	0.1855
0.6452	28.1195	48000	0.0830	0.1882
0.6682	28.7053	49000	0.0828	0.1863
0.6357	29.2912	50000	0.0829	0.1877
0.6645	29.8770	51000	0.0831	0.1898

Framework versions

Transformers 4.51.3
Pytorch 2.6.0+cu124
Datasets 3.6.0
Tokenizers 0.21.1

Loading…

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: 30

mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.1291	0.5858	1000	0.0912	0.1978
1.7057	1.1716	2000	0.0912	0.2003
1.7162	1.7575	3000	0.0912	0.2060
1.4996	2.3433	4000	0.0901	0.2047
1.3942	2.9291	5000	0.0883	0.1951
1.2285	3.5149	6000	0.0876	0.1957
1.0637	4.1008	7000	0.0873	0.1920
1.1144	4.6866	8000	0.0865	0.1927
1.0164	5.2724	9000	0.0858	0.1923
0.9812	5.8582	10000	0.0856	0.1941
0.8927	6.4441	11000	0.0849	0.2017
0.8936	7.0299	12000	0.0844	0.1961
0.8718	7.6157	13000	0.0854	0.1979
0.9019	8.2015	14000	0.0847	0.1854
0.8293	8.7873	15000	0.0847	0.1983
0.8363	9.3732	16000	0.0842	0.1982
0.8034	9.9590	17000	0.0840	0.1975
0.8462	10.5448	18000	0.0855	0.1953
0.8824	11.1306	19000	0.0848	0.1930
0.8591	11.7165	20000	0.0849	0.1838
0.8339	12.3023	21000	0.0842	0.1863
0.8573	12.8881	22000	0.0836	0.1926
0.7445	13.4739	23000	0.0839	0.1842
0.783	14.0598	24000	0.0836	0.1842
0.7263	14.6456	25000	0.0839	0.1824
0.7634	15.2314	26000	0.0835	0.1826
0.7379	15.8172	27000	0.0834	0.1829
0.7902	16.4030	28000	0.0842	0.1811
0.8261	16.9889	29000	0.0841	0.1849
0.7531	17.5747	30000	0.0840	0.1867
0.7166	18.1605	31000	0.0839	0.1905
0.7976	18.7463	32000	0.0841	0.1838
0.7008	19.3322	33000	0.0835	0.1864
0.707	19.9180	34000	0.0833	0.1872
0.6865	20.5038	35000	0.0835	0.1844
0.6927	21.0896	36000	0.0834	0.1882
0.7014	21.6755	37000	0.0835	0.1861
0.6951	22.2613	38000	0.0833	0.1874
0.6848	22.8471	39000	0.0834	0.1927
0.7096	23.4329	40000	0.0834	0.1936
0.6952	24.0187	41000	0.0835	0.1933
0.692	24.6046	42000	0.0833	0.1930
0.6552	25.1904	43000	0.0831	0.1867
0.6641	25.7762	44000	0.0832	0.1874
0.6921	26.3620	45000	0.0833	0.1880
0.6894	26.9479	46000	0.0832	0.1855
0.7041	27.5337	47000	0.0827	0.1855
0.6452	28.1195	48000	0.0830	0.1882
0.6682	28.7053	49000	0.0828	0.1863
0.6357	29.2912	50000	0.0829	0.1877
0.6645	29.8770	51000	0.0831	0.1898

Framework versions

Transformers 4.51.3

Pytorch 2.6.0+cu124

Datasets 3.6.0

Tokenizers 0.21.1

Baselhany/Distilation_Whisper_base_CKP

Whisper base AR - BA

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

Whisper base AR - BA

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions