whisper-large-v3-sandi-train-dev-1-transcript

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

• Loss: 1.1193
• Wer: 166.8810
• Cer: 105.7146
• Decode Runtime: 549.2811
• Wer Runtime: 0.2378
• Cer Runtime: 0.4776

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
• gradient_accumulation_steps: 32
• total_train_batch_size: 1024
• 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: 28

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Decode Runtime	Wer Runtime	Cer Runtime
2.3646	1.0357	7	1.6548	221.5703	172.3091	535.1102	0.2869	0.5693
1.4099	2.0714	14	1.3247	186.9695	110.7974	535.4955	0.2481	0.4912
1.2041	3.1071	21	1.1672	153.4217	104.1363	554.2174	0.2334	0.4790
1.1115	4.1429	28	1.1193	166.8810	105.7146	549.2811	0.2378	0.4776

Framework versions

• PEFT 0.15.2
• Transformers 4.51.3
• Pytorch 2.4.1+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1