Tirendaz/whisper-large-v3-turbo

Whisper Large V3 Pro on the Turkish Dataset - Evren Ozkip

This model is a fine-tuned version of openai/whisper-small on the Turkish Common Voice 13 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4715
• Wer Ortho: 35.2262
• Wer: 31.3503

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
• 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: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• training_steps: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
No log	0.08	20	0.3358	25.0241	21.7186
0.4294	0.16	40	0.3339	26.0828	22.7573
0.2121	0.24	60	0.5079	54.9567	50.1416
0.3798	0.32	80	0.4675	55.2454	50.8026
0.3469	0.4	100	0.4715	35.2262	31.3503

Framework versions

• Transformers 4.51.3
• Pytorch 2.1.1+cu121
• Datasets 3.6.0
• Tokenizers 0.21.1