ntnu-smil/whisper-large-v3-turbo-ami-1-merged

whisper-large-v3-turbo-ami-1

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the ntnu-smil/ami-1s-ft dataset. It achieves the following results on the evaluation set:

• Loss: 3.4985
• Wer: 40.4100
• Cer: 34.3391
• Decode Runtime: 0.0102
• Wer Runtime: 0.0078
• Cer Runtime: 0.0077

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-06
• train_batch_size: 32
• eval_batch_size: 128
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• training_steps: 2000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Decode Runtime	Wer Runtime	Cer Runtime
No log	0	0	3.9428	53.6237	39.2308	0.0108	0.0079	0.0071
0.4259	0.1	200	3.3403	36.4934	30.1718	0.0097	0.0080	0.0074
0.1836	0.2	400	3.3032	37.2255	32.0015	0.0097	0.0079	0.0072
0.333	0.3	600	3.3810	38.5798	32.6363	0.0110	0.0084	0.0072
0.1716	0.4	800	3.4782	39.2020	33.0769	0.0096	0.0081	0.0069
0.1919	0.5	1000	3.4548	39.1654	33.6146	0.0097	0.0079	0.0071
0.2053	0.6	1200	3.4784	39.6413	33.5848	0.0101	0.0084	0.0072
0.126	0.7	1400	3.5044	40.3734	34.0553	0.0100	0.0083	0.0073
0.331	0.8	1600	3.4866	40.0439	34.0777	0.0100	0.0084	0.0072
0.1882	0.9	1800	3.4982	40.0805	34.1001	0.0100	0.0079	0.0075
0.1013	1.0	2000	3.4985	40.4100	34.3391	0.0102	0.0078	0.0077

Framework versions

• PEFT 0.17.0
• Transformers 4.54.1
• Pytorch 2.8.0.dev20250319+cu128
• Datasets 3.6.0
• Tokenizers 0.21.4