whisper_base_test

This model is a fine-tuned version of openai/whisper-base on the AIhub_foreign_dataset4 dataset. It achieves the following results on the evaluation set:

• Loss: 0.9395
• Cer: 48.1817

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: 2e-05
• train_batch_size: 16
• eval_batch_size: 8
• 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: cosine
• lr_scheduler_warmup_steps: 500
• training_steps: 6000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
1.0479	0.7299	1000	0.9942	53.5695
0.983	1.4599	2000	0.9553	50.5824
0.7329	2.1898	3000	0.9432	56.3785
0.7979	2.9197	4000	0.9299	48.6001
0.6956	3.6496	5000	0.9368	47.2723
0.5971	4.3796	6000	0.9395	48.1817

Framework versions

• Transformers 4.51.0.dev0
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1