Baselhany/Distilation_Whisper_base_CKP_256

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: 0.0853
• Wer: 0.1969

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: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.0599	0.5858	1000	0.0910	0.2075
1.6156	1.1716	2000	0.0921	0.1917
1.5706	1.7575	3000	0.0891	0.1953
1.3401	2.3433	4000	0.0880	0.1882
1.2238	2.9291	5000	0.0865	0.1886
1.0654	3.5149	6000	0.0860	0.1922
1.0904	4.1008	7000	0.0859	0.2000
1.2607	4.6866	8000	0.0872	0.1882
1.147	5.2724	9000	0.0870	0.1944
1.1237	5.8582	10000	0.0856	0.1905
1.0093	6.4441	11000	0.0849	0.2001
0.9993	7.0299	12000	0.0839	0.1888
0.8718	7.6157	13000	0.0844	0.1894
0.8877	8.2015	14000	0.0838	0.1908
0.8187	8.7873	15000	0.0843	0.1957
0.8235	9.3732	16000	0.0838	0.1975
0.7972	9.9590	17000	0.0835	0.1911
0.8203	10.5448	18000	0.0844	0.1866
0.8593	11.1306	19000	0.0843	0.1916
0.8279	11.7165	20000	0.0840	0.1905
0.806	12.3023	21000	0.0827	0.1897
0.8343	12.8881	22000	0.0832	0.1891
0.7252	13.4739	23000	0.0830	0.1845
0.7685	14.0598	24000	0.0830	0.1919
0.7085	14.6456	25000	0.0829	0.1975

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1