ntnu-smil/whisper-large-v3-turbo-sandi-train-dev-1-rich-transcript-32-2x-merged

whisper-large-v3-turbo-sandi-train-dev-1-rich-transcript-32-2x

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: 0.6974
• Wer: 21.6454
• Cer: 15.6079
• Decode Runtime: 219.4803
• Wer Runtime: 0.1696
• Cer Runtime: 0.3471

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
0.8635	0.1667	122	0.8126	34.2357	24.0646	224.8237	0.1969	0.3499
0.7317	1.0546	244	0.7678	24.2432	17.5239	221.5193	0.2016	0.3534
1.1393	1.2213	366	0.7359	22.1059	15.8230	223.8618	0.1919	0.3583
1.0918	2.1093	488	0.7150	21.2576	15.2688	223.7214	0.1872	0.3538
0.686	2.2760	610	0.7036	22.0370	15.8432	206.2499	0.1890	0.3564
0.7399	3.1639	732	0.6974	21.6454	15.6079	219.4803	0.1696	0.3471

Framework versions

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