vit-lr-cosine-warmup

This model is a fine-tuned version of google/vit-base-patch16-224 on the skin-cancer dataset. It achieves the following results on the evaluation set:

• Loss: 0.4736
• Accuracy: 0.8395
• Precision: 0.8318
• Recall: 0.8395
• F1: 0.8308

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: 0.0001
• 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: cosine
• lr_scheduler_warmup_steps: 770
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
0.86	1.0	321	0.5250	0.8141	0.8100	0.8141	0.8011
0.4517	2.0	642	0.5117	0.8221	0.8347	0.8221	0.8100
0.3512	3.0	963	0.4736	0.8395	0.8318	0.8395	0.8308
0.2184	4.0	1284	0.4797	0.8568	0.8536	0.8568	0.8505
0.1264	5.0	1605	0.6212	0.8547	0.8552	0.8547	0.8530
0.0687	6.0	1926	0.7659	0.8464	0.8476	0.8464	0.8402
0.0463	7.0	2247	0.8237	0.8519	0.8546	0.8519	0.8469
0.0373	8.0	2568	0.8712	0.8377	0.8493	0.8377	0.8415
0.0347	9.0	2889	0.8181	0.8568	0.8550	0.8568	0.8534
0.0263	10.0	3210	1.0705	0.8447	0.8389	0.8447	0.8308
0.0289	11.0	3531	0.9376	0.8589	0.8606	0.8589	0.8550
0.0164	12.0	3852	0.9714	0.8634	0.8611	0.8634	0.8611
0.0077	13.0	4173	1.2992	0.8398	0.8396	0.8398	0.8243

Framework versions

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