whisper-tiny-kor-16k-hf-ep100

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

• Loss: 0.9175
• Cer: 29.8227

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: 5e-05
• train_batch_size: 128
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.1166	7.94	1000	0.6351	34.2349
0.0024	15.87	2000	0.7071	33.1900
0.0015	23.81	3000	0.7484	29.1958
0.0009	31.75	4000	0.7972	29.5301
0.0008	39.68	5000	0.8082	29.1122
0.0003	47.62	6000	0.8581	29.5182
0.0	55.56	7000	0.8756	29.6316
0.0	63.49	8000	0.8922	29.6316
0.0004	71.43	9000	0.8693	29.5958
0.0	79.37	10000	0.8930	29.9003
0.0	87.3	11000	0.9081	29.7212
0.0	95.24	12000	0.9175	29.8227

Framework versions

• Transformers 4.36.2
• Pytorch 2.0.1+cu117
• Datasets 2.16.1
• Tokenizers 0.15.0