whisper-large-v3-sandi-7k-64-448steps

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: 0.5722
• Wer: 24.0947
• Cer: 56.2387
• Decode Runtime: 203.1841
• Wer Runtime: 0.1735
• Cer Runtime: 0.3276

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: 2
• total_train_batch_size: 64
• 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: 448

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Decode Runtime	Wer Runtime	Cer Runtime
0.6419	1.0223	112	0.6663	20.0083	24.8032	187.4701	0.1653	0.2986
0.6651	2.0446	224	0.6117	20.0564	34.0018	189.8527	0.1717	0.3134
0.4682	3.0670	336	0.5826	21.0683	32.6385	190.6981	0.1750	0.3082
0.8059	4.0893	448	0.5722	24.0947	56.2387	203.1841	0.1735	0.3276

Framework versions

• PEFT 0.15.2
• Transformers 4.51.3
• Pytorch 2.4.1+cu124
• Datasets 3.5.1
• Tokenizers 0.21.1