ntnu-smil/whisper-large-v3-sandi-3k-1024-28steps-merged

whisper-large-v3-sandi-3k-1024-28steps

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: 1.0370
• Wer: 78.0238
• Cer: 215.7449
• Decode Runtime: 252.2954
• Wer Runtime: 0.1988
• Cer Runtime: 0.4668

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
2.6493	2.0357	7	1.3706	67.2148	209.8014	252.1421	0.2012	0.4813
1.1778	4.0714	14	1.1881	82.9771	226.7708	259.9550	0.1999	0.4853
0.9983	6.1071	21	1.0717	79.1953	220.4455	259.7244	0.2083	0.4860
1.9008	9.0357	28	1.0370	78.0238	215.7449	252.2954	0.1988	0.4668

Framework versions

• PEFT 0.15.2
• Transformers 4.48.2
• Pytorch 2.4.1+cu124
• Datasets 3.5.1
• Tokenizers 0.21.1