whisper-large-v3-turbo-sandi-train-1-rich-transcript-32

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the ntnu-smil/sandi2025-ds dataset. It achieves the following results on the evaluation set:

• Loss: 0.7920
• Wer: 23.2484
• Cer: 16.5856
• Decode Runtime: 203.4514
• Wer Runtime: 0.1639
• Cer Runtime: 0.3157

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: 7e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.98) and epsilon=1e-06 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• training_steps: 732

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Decode Runtime	Wer Runtime	Cer Runtime
0.9653	0.1667	122	0.8334	103.3411	64.3972	245.8729	0.1967	0.3575
1.1313	1.1667	244	0.8073	53.0086	33.9467	210.6469	0.1851	0.3293
0.54	2.1667	366	0.7915	25.4142	18.3008	196.4910	0.1906	0.3139
0.3761	3.1667	488	0.7882	24.2463	17.3425	196.9004	0.1675	0.3169
0.8462	4.1667	610	0.7921	23.4051	16.7178	197.5723	0.1661	0.3141
0.9957	5.1667	732	0.7920	23.2484	16.5856	203.4514	0.1639	0.3157

Framework versions

• PEFT 0.15.2
• Transformers 4.52.2
• Pytorch 2.8.0.dev20250319+cu128
• Datasets 3.6.0
• Tokenizers 0.21.1