gwanju_large2_model

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

• Loss: 0.3321
• Cer: 438.5339

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: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.4681	0.0741	250	0.4882	92.4888
0.4609	0.1482	500	0.4507	180.4507
0.4749	0.2223	750	0.4351	148.4249
0.4248	0.2964	1000	0.4260	50.0864
0.4433	0.3705	1250	0.3998	107.5518
0.3667	0.4446	1500	0.3907	296.2817
0.3805	0.5187	1750	0.3795	308.2578
0.3571	0.5928	2000	0.3770	396.0998
0.4312	0.6669	2250	0.3644	470.9584
0.3445	0.7410	2500	0.3562	392.7995
0.4036	0.8151	2750	0.3485	468.5345
0.3523	0.8892	3000	0.3426	459.9051
0.3541	0.9632	3250	0.3377	456.2648
0.2252	1.0373	3500	0.3343	450.6082
0.2063	1.1114	3750	0.3333	444.6852
0.2018	1.1855	4000	0.3321	438.5339

Framework versions

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