korean-small_t36

This model is a fine-tuned version of openai/whisper-small on the korean_samll_dataset13 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2805
• Cer: 10.8675

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: 16
• eval_batch_size: 8
• 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: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.3938	0.11	200	0.3982	14.8401
0.3479	0.21	400	0.3561	12.9310
0.3206	0.32	600	0.3334	12.3517
0.3094	0.42	800	0.3222	12.0216
0.3088	0.53	1000	0.3120	11.6705
0.2792	0.63	1200	0.3058	11.9337
0.2877	0.74	1400	0.2988	11.9042
0.2722	0.84	1600	0.2913	11.6501
0.285	0.95	1800	0.2881	11.5122
0.1822	1.05	2000	0.2870	12.0730
0.1829	1.16	2200	0.2861	11.0178
0.1843	1.26	2400	0.2850	11.4228
0.1869	1.37	2600	0.2844	11.1706
0.1886	1.47	2800	0.2826	11.0313
0.1816	1.58	3000	0.2805	10.8675
0.1828	1.69	3200	0.2792	11.0108
0.1844	1.79	3400	0.2774	10.9839
0.1847	1.9	3600	0.2747	11.2211
0.1759	2.0	3800	0.2742	11.2830
0.112	2.11	4000	0.2814	11.5537
0.1185	2.21	4200	0.2825	10.9629
0.1142	2.32	4400	0.2812	11.4553
0.1079	2.42	4600	0.2812	11.3894
0.1139	2.53	4800	0.2811	11.0738
0.1085	2.63	5000	0.2811	11.3989
0.1096	2.74	5200	0.2807	11.0138
0.1087	2.84	5400	0.2804	11.1387
0.1103	2.95	5600	0.2801	11.1097

Framework versions

• Transformers 4.39.0.dev0
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2