whisper-finetuned-v3_15e_augment_new

This model is a fine-tuned version of openai/whisper-large-v3-turbo on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1010
• Wer: 52.2851
• Cer: 27.6080

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Use 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: 500
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.102	1.0	1771	0.0968	65.6548	32.1205
0.0639	2.0	3542	0.0787	60.3001	29.2997
0.0398	3.0	5313	0.0719	58.4243	28.8596
0.0292	4.0	7084	0.0773	56.4120	29.1689
0.017	5.0	8855	0.0811	56.5484	28.6147
0.0124	6.0	10626	0.0796	56.3779	28.3262
0.0107	7.0	12397	0.0851	55.4911	28.6064
0.0066	8.0	14168	0.0794	54.9795	27.8405
0.006	9.0	15939	0.0928	53.7517	27.8633
0.0025	10.0	17710	0.0934	54.1610	28.0605
0.0028	11.0	19481	0.0935	53.6153	27.7139
0.0013	12.0	21252	0.0956	52.7626	27.9007
0.0026	13.0	23023	0.0964	53.1037	27.6890
0.0003	14.0	24794	0.1011	52.7285	27.7284
0.0001	14.9918	26550	0.1010	52.2851	27.6080

Framework versions

• Transformers 4.51.3
• Pytorch 2.7.0+cu126
• Datasets 3.6.0
• Tokenizers 0.21.1