notebotIE/whisper-large-v2-swiss-german

Whisper Large V2 - Swiss German

This model is a fine-tuned version of openai/whisper-large-v2 on the SwissDialDataset_ETH dataset. It achieves the following results on the evaluation set:

• Loss: 0.2462
• Wer Ortho: 0.2459
• Wer: 0.1577
• Cer: 0.0373

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: 4
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 5
• training_steps: 250
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer	Cer
0.4177	0.2460	50	0.3617	0.3915	0.3244	0.1232
0.285	0.4920	100	0.3100	0.2905	0.2013	0.0409
0.2659	0.7380	150	0.2632	0.3753	0.2909	0.4770
0.2401	0.9840	200	0.2372	0.2541	0.1568	0.0321
0.1192	1.2300	250	0.2462	0.2459	0.1577	0.0373

Framework versions

• Transformers 4.46.3
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.20.3