whisper_small_test

This model is a fine-tuned version of openai/whisper-small on the AIhub_KoSpeech_dataset2 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3283
• Cer: 10.1965

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• 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
• training_steps: 7000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.7295	0.0790	500	0.7146	24.9317
0.6543	0.1580	1000	0.6799	48.0012
0.6291	0.2370	1500	0.6113	18.1741
0.577	0.3161	2000	0.5714	18.5980
0.4792	0.3951	2500	0.5176	15.9901
0.4775	0.4741	3000	0.4835	15.4528
0.4358	0.5531	3500	0.4543	15.3170
0.425	0.6321	4000	0.4246	13.0461
0.3961	0.7111	4500	0.4018	12.9581
0.3776	0.7901	5000	0.3768	11.7971
0.3923	0.8692	5500	0.3601	11.4129
0.3601	0.9482	6000	0.3406	10.8236
0.2029	1.0272	6500	0.3321	10.3529
0.1899	1.1062	7000	0.3283	10.1965

Framework versions

• Transformers 4.52.0.dev0
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1