lalok/gyeongsan_address_firestation_ko_14000hr_50t

This model is a fine-tuned version of openai/whisper-medium on the lalok/gyeongsan_address_firestation_ko_14000hr dataset. It achieves the following results on the evaluation set:

• Loss: 0.2038
• Cer: 13.6895

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: 100
• training_steps: 50000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.286	0.0908	5000	0.2856	16.6135
0.2355	0.1816	10000	0.2655	15.5819
0.2454	0.2724	15000	0.2538	14.9019
0.2073	0.3632	20000	0.2423	14.3891
0.2314	0.4540	25000	0.2330	14.2384
0.218	0.5449	30000	0.2260	14.5535
0.2124	0.6357	35000	0.2181	13.7738
0.2182	0.7265	40000	0.2118	14.0623
0.2166	0.8173	45000	0.2067	13.9199
0.2069	0.9081	50000	0.2038	13.6895

Framework versions

• Transformers 4.41.2
• Pytorch 2.2.2+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1