whisper-medium-sundanese

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: 0.0250
• Wer: 2.9531

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: 4
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• 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
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0901	0.2286	500	0.0849	9.8180
0.0475	0.4571	1000	0.0526	5.0729
0.0332	0.6857	1500	0.0390	9.4872
0.0394	0.9143	2000	0.0349	3.9578
0.0066	1.1426	2500	0.0303	16.5452
0.0122	1.3712	3000	0.0280	7.1468
0.0088	1.5998	3500	0.0275	2.4537
0.0053	1.8283	4000	0.0262	2.8857
0.0044	2.0567	4500	0.0253	4.2672
0.0017	2.2853	5000	0.0250	2.9531

Framework versions

• Transformers 4.57.3
• Pytorch 2.4.1+cu124
• Datasets 2.21.0
• Tokenizers 0.22.1