whisper-medium-bemgen-100f50m-model

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: 0.4354
• Wer: 0.3898

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: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Use 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
3.1512	0.2641	200	0.8282	0.6534
2.4008	0.5282	400	0.6203	0.5097
2.2459	0.7923	600	0.5506	0.4644
1.3319	1.0555	800	0.5029	0.4017
1.5588	1.3196	1000	0.4675	0.3897
1.2908	1.5837	1200	0.4534	0.3727
1.4258	1.8478	1400	0.4354	0.3898
0.6383	2.1109	1600	0.4480	0.3600
0.6079	2.3750	1800	0.4444	0.3482
0.5709	2.6392	2000	0.4367	0.3405

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0