whisper-40hrs-classifier

This model is a fine-tuned version of openai/whisper-large-v2 on the JASMIN-CGN dataset. It achieves the following results on the evaluation set:

• Loss: 0.3795
• Wer: 19.7571

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: 1e-05
• train_batch_size: 48
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 106
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.9643	0.1416	50	1.2066	37.9676
1.9513	0.2833	100	1.1128	37.2228
1.8194	0.4249	150	0.9154	36.0486
1.7286	0.5666	200	0.6881	34.2369
1.5144	0.7082	250	0.5460	29.6038
1.4289	0.8499	300	0.4679	23.3871
1.3439	0.9915	350	0.4419	19.7370
1.4136	1.1331	400	0.4268	20.6462
1.357	1.2748	450	0.4156	19.9919
1.3042	1.4164	500	0.4064	19.3075
1.2603	1.5581	550	0.3997	18.7104
1.2128	1.6997	600	0.3964	18.6433
1.3441	1.8414	650	0.3957	20.0289
1.3065	1.9830	700	0.3894	19.0861
1.2867	2.1246	750	0.3886	19.7739
1.2856	2.2663	800	0.3867	20.4583
1.2355	2.4079	850	0.3823	19.2673
1.2697	2.5496	900	0.3808	19.4585
1.2894	2.6912	950	0.3816	20.3543
1.2832	2.8329	1000	0.3806	19.7638
1.2718	2.9745	1050	0.3795	19.7571

Framework versions

• PEFT 0.16.0
• Transformers 4.52.0
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.2