Whisper-Base-MN-EG - Mina Nasser

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

• Loss: 0.5446
• Wer: 49.9256
• Cer: 28.5219

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: 6
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.1391	1.0989	500	0.5976	57.8994	33.4040
0.8563	2.1978	1000	0.5692	52.5482	30.5180
0.7231	3.2967	1500	0.5594	53.2405	31.4875
0.6320	4.3956	2000	0.5600	49.4351	27.8913
0.6085	5.4945	2500	0.5604	50.2102	28.6843

Framework versions

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