whisper-large-v3-sandi-train-dev-6

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

• Loss: 0.6141
• Wer: 195.1740
• Cer: 103.8099
• Decode Runtime: 302.6853
• Wer Runtime: 0.1962
• Cer Runtime: 0.3691

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: 5e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Use 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: 392

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Decode Runtime	Wer Runtime	Cer Runtime
0.5955	1.0	56	0.7708	199.7404	87.4116	297.6150	0.1968	0.3600
0.583	2.0	112	0.6938	198.2328	86.9214	301.9292	0.1913	0.3584
0.4582	3.0	168	0.6606	197.2721	89.4544	308.3536	0.1964	0.3621
0.5814	4.0	224	0.6398	196.7833	95.2803	300.6860	0.1962	0.3622
0.4986	5.0	280	0.6255	196.2085	95.9717	302.4188	0.1945	0.3630
0.4225	6.0	336	0.6170	195.3584	103.5412	301.1984	0.1943	0.3702
0.465	7.0	392	0.6141	195.1740	103.8099	302.6853	0.1962	0.3691

Framework versions

• PEFT 0.15.1
• Transformers 4.48.3
• Pytorch 2.6.0
• Datasets 3.5.0
• Tokenizers 0.21.1