vit-weight-decay-1e-3

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: 0.5183
• Accuracy: 0.8381
• Precision: 0.8396
• Recall: 0.8381
• F1: 0.8356

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: 1733
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
1.7663	1.0	321	0.9397	0.6900	0.6694	0.6900	0.6516
1.173	2.0	642	0.6933	0.7257	0.7710	0.7257	0.7360
1.035	3.0	963	0.6482	0.7611	0.7475	0.7611	0.7234
0.9945	4.0	1284	0.6027	0.7625	0.7872	0.7625	0.7661
0.9338	5.0	1605	0.6132	0.7580	0.7989	0.7580	0.7709
0.983	6.0	1926	0.6297	0.7483	0.7977	0.7483	0.7652
0.8867	7.0	2247	0.7642	0.7021	0.7986	0.7021	0.7286
0.8273	8.0	2568	0.6594	0.7386	0.8000	0.7386	0.7546
0.7984	9.0	2889	0.5539	0.7989	0.8082	0.7989	0.7986
0.7191	10.0	3210	0.5732	0.7715	0.8286	0.7715	0.7848
0.6859	11.0	3531	0.5409	0.8131	0.8382	0.8131	0.8164
0.5944	12.0	3852	0.6452	0.7968	0.8273	0.7968	0.8065
0.5064	13.0	4173	0.5183	0.8381	0.8396	0.8381	0.8356
0.4759	14.0	4494	0.6007	0.8145	0.8394	0.8145	0.8216
0.4203	15.0	4815	0.5580	0.8200	0.8412	0.8200	0.8263
0.3931	16.0	5136	0.6186	0.8027	0.8415	0.8027	0.8145
0.3076	17.0	5457	0.5484	0.8336	0.8531	0.8336	0.8395
0.2729	18.0	5778	0.5892	0.8419	0.8495	0.8419	0.8447
0.2265	19.0	6099	0.6573	0.8325	0.8543	0.8325	0.8386
0.2685	20.0	6420	0.6374	0.8488	0.8577	0.8488	0.8517
0.1935	21.0	6741	0.6543	0.8568	0.8559	0.8568	0.8507
0.1486	22.0	7062	0.6030	0.8665	0.8671	0.8665	0.8657
0.1628	23.0	7383	0.6315	0.8717	0.8717	0.8717	0.8676

Framework versions

• Transformers 4.40.0.dev0
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2