whisper-large-v3-ami-1

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

• Loss: 3.6457
• Wer: 73.2830
• Cer: 65.1890
• Decode Runtime: 3.7197
• Wer Runtime: 0.0090
• Cer Runtime: 0.0152

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: 128
• eval_batch_size: 128
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 1024
• 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
• lr_scheduler_warmup_steps: 50
• training_steps: 130

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Decode Runtime	Wer Runtime	Cer Runtime
2.2365	0.0769	10	3.2101	71.2225	305.1720	5.7416	0.0099	0.0322
1.9464	0.1538	20	3.1678	81.2843	319.6875	5.8313	0.0098	0.0337
1.5994	0.2308	30	3.0765	106.4904	341.3692	5.8220	0.0105	0.0351
1.1357	0.3077	40	3.2982	129.5330	214.6070	5.6144	0.0102	0.0259
0.4404	0.3846	50	3.4638	72.2871	98.6465	3.8830	0.0093	0.0179
0.3252	0.4615	60	3.3927	65.1099	80.9729	3.7645	0.0091	0.0167
0.3713	1.0231	70	3.4800	58.9629	49.3854	3.4950	0.0090	0.0142
0.2562	1.1	80	3.5965	54.0522	31.3522	3.3013	0.0089	0.0130
0.1821	1.1769	90	3.6241	70.4327	56.6693	3.6241	0.0089	0.0146
0.1847	1.2538	100	3.6725	66.2775	50.4512	3.6175	0.0090	0.2387
0.2257	1.3308	110	3.6518	64.8695	50.6408	3.5330	0.0090	0.0141
0.2672	1.4077	120	3.6463	69.7802	59.8928	3.6917	0.0090	0.0146
0.2578	1.4846	130	3.6457	73.2830	65.1890	3.7197	0.0090	0.0152

Framework versions

• PEFT 0.14.0
• Transformers 4.48.0
• Pytorch 2.5.1
• Datasets 3.2.0
• Tokenizers 0.21.0