basha3la2/whisper-base-de

Whisper base de - Alaa Albasha

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

• Loss: 0.2420
• Wer: 14.3637

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: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.2115	0.0651	1000	0.2977	17.2904
1.0521	0.1301	2000	0.2640	15.5865
0.9439	0.1952	3000	0.2487	14.7318
0.9417	0.2602	4000	0.2420	14.3637

Framework versions

• Transformers 5.15.1
• Pytorch 2.10.0+cu128
• Datasets 5.0.1
• Tokenizers 0.22.2