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

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

This model is a fine-tuned version of ntnu-smil/whisper-large-v3-turbo-sandi-train-1-ex-transcript-32-merged on the ntnu-smil/sandi2025-ds dataset. It achieves the following results on the evaluation set:

• Loss: 0.6989
• Wer: 36.0133
• Cer: 26.8844
• Decode Runtime: 231.3002
• Wer Runtime: 0.1747
• Cer Runtime: 0.3428

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
0.8381	0.1667	122	0.8004	38.2546	28.4316	221.0426	0.1936	0.3718
0.721	1.0546	244	0.7636	29.8239	21.6741	213.1620	0.1798	0.3544
1.1003	1.2213	366	0.7355	36.6304	27.2175	235.8508	0.1728	0.3398
1.07	2.1093	488	0.7148	32.8920	24.3094	232.6055	0.1917	0.3679
0.6578	2.2760	610	0.7043	35.3328	26.2606	232.2468	0.1869	0.3647
0.7273	3.1639	732	0.6989	36.0133	26.8844	231.3002	0.1747	0.3428

Framework versions

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