Whisper-Small-MN - Mina Nasser

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

• Loss: 0.3650
• Wer: 37.8043
• Cer: 22.2604

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: 2
• total_train_batch_size: 32
• 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: cosine
• lr_scheduler_warmup_steps: 450
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.8504	0.5495	250	0.4142	40.0875	23.4493
0.6431	1.0989	500	0.3914	39.1001	23.3363
0.6066	1.6484	750	0.3828	40.6354	24.3984
0.4221	2.1978	1000	0.3848	38.0043	22.2810
0.4463	2.7473	1250	0.3800	37.7134	22.3369
0.3035	3.2967	1500	0.3901	37.7368	22.3377
0.3085	3.8462	1750	0.3890	37.8684	22.5591

Framework versions

• Transformers 5.2.0
• Pytorch 2.10.0+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2