rmacek/whisper-small-de

Whisper Small ZIB2

This model is a fine-tuned version of openai/whisper-small on the ZIB2 Common Voice dataset. It achieves the following results on the evaluation set:

• Loss: 0.3366
• Wer: 28.9384

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: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2391	10.0	100	0.2837	33.5616
0.0035	20.0	200	0.2701	27.7397
0.0012	30.0	300	0.2847	27.5685
0.0006	40.0	400	0.2990	27.9110
0.0004	50.0	500	0.3118	28.5959
0.0003	60.0	600	0.3221	28.5959
0.0002	70.0	700	0.3287	28.7671
0.0002	80.0	800	0.3333	28.9384
0.0002	90.0	900	0.3357	28.9384
0.0002	100.0	1000	0.3366	28.9384

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2