whisper-small-ko-Y_freq_speed

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

• Loss: 0.2779
• Cer: 10.3383

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.5211	0.13	100	0.3156	7.5775
0.3128	0.26	200	0.2816	8.0651
0.2468	0.39	300	0.2944	8.4998
0.2386	0.52	400	0.2764	7.9182
0.2188	0.64	500	0.2806	8.8992
0.195	0.77	600	0.2818	8.1473
0.2067	0.9	700	0.2759	8.6114
0.094	1.03	800	0.2725	8.3940
0.0733	1.16	900	0.2766	8.8170
0.0864	1.29	1000	0.2839	8.6701
0.0831	1.42	1100	0.2810	10.7848
0.0874	1.55	1200	0.2781	11.1078
0.0689	1.68	1300	0.2766	10.0329
0.0752	1.81	1400	0.2771	10.4206
0.0777	1.93	1500	0.2779	10.3383

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.0+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0