whisper-medium-bemgen-combined-52

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

• Loss: 0.4951
• Wer: 0.4190

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: 2
• eval_batch_size: 2
• seed: 52
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• 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: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.9356	0.2859	200	0.7852	0.6652
0.625	0.5718	400	0.6335	0.5398
0.5464	0.8578	600	0.5583	0.4753
0.3677	1.1430	800	0.5201	0.4392
0.3569	1.4289	1000	0.5456	0.4740
0.3603	1.7148	1200	0.5105	0.4354
0.3474	2.0	1400	0.4951	0.4190
0.1719	2.2859	1600	0.5337	0.4347
0.2137	2.5718	1800	0.5334	0.4440
0.1976	2.8578	2000	0.5259	0.4206

Framework versions

• Transformers 4.53.0.dev0
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.0