whisper-medium-tarteel-quraan

This model is a fine-tuned version of openai/whisper-medium on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0112
• Cer: 0.0736
• Wer: 0.1110

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
• 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: 100
• training_steps: 2000

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
0.0737	0.1145	125	0.0708	0.1345	0.3640
0.0585	0.2289	250	0.0494	0.1201	0.2884
0.0399	0.3434	375	0.0418	0.0987	0.2219
0.0262	0.4579	500	0.0372	0.0946	0.2043
0.0269	0.5723	625	0.0328	0.0903	0.1869
0.0415	0.6868	750	0.0306	0.0895	0.1774
0.0287	0.8013	875	0.0268	0.0871	0.1635
0.0246	0.9158	1000	0.0234	0.0823	0.1520
0.0160	1.0302	1125	0.0228	0.0824	0.1474
0.0190	1.1447	1250	0.0204	0.0782	0.1356
0.0176	1.2592	1375	0.0192	0.0797	0.1368
0.0091	1.3736	1500	0.0170	0.0774	0.1284
0.0123	1.4881	1625	0.0155	0.0768	0.1271
0.0080	1.6026	1750	0.0134	0.0746	0.1175
0.0055	1.7170	1875	0.0121	0.0748	0.1154
0.0080	1.8315	2000	0.0112	0.0736	0.1110

Framework versions

• Transformers 5.1.0
• Pytorch 2.4.1+cu124
• Datasets 2.20.0
• Tokenizers 0.22.2