whisper-medium-bemgen-combined-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: 0.4200
• Wer: 0.3332

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

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.7922	0.1980	200	0.8807	0.6663
1.3451	0.3960	400	0.6738	0.5310
1.1869	0.5941	600	0.5800	0.4613
0.9659	0.7921	800	0.5199	0.4211
0.8946	0.9901	1000	0.4816	0.3967
0.6349	1.1881	1200	0.4725	0.3726
0.6238	1.3861	1400	0.4549	0.3603
0.6244	1.5842	1600	0.4495	0.3648
0.5724	1.7822	1800	0.4362	0.3451
0.6594	1.9802	2000	0.4200	0.3332
0.3207	2.1782	2200	0.4395	0.3353
0.301	2.3762	2400	0.4479	0.3275
0.2863	2.5743	2600	0.4369	0.3358

Framework versions

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