ntnu-smil/whisper-large-v3-turbo-sandi-train-1-ex-transcript-32-r128-merged

whisper-large-v3-turbo-sandi-train-1-ex-transcript-32-r128

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: 0.8115
• Wer: 29.3172
• Cer: 21.2742
• Decode Runtime: 200.5385
• Wer Runtime: 0.1825
• Cer Runtime: 0.3207

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
0.9816	0.1667	122	0.8325	52.6136	40.1293	213.3993	0.1959	0.3519
1.12	1.1667	244	0.8067	41.8140	31.3145	206.7493	0.1829	0.3292
0.5435	2.1667	366	0.7957	28.9535	21.0430	196.2685	0.1781	0.3176
0.3301	3.1667	488	0.7980	28.7450	20.8476	197.7640	0.1800	0.3185
0.8564	4.1667	610	0.8075	30.0818	21.8933	200.7786	0.1811	0.3254
0.8796	5.1667	732	0.8115	29.3172	21.2742	200.5385	0.1825	0.3207

Framework versions

• PEFT 0.15.2
• Transformers 4.48.2
• Pytorch 2.4.1+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1