vit-epsilon-1e-9

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: 0.5143
• Accuracy: 0.8131
• Precision: 0.8359
• Recall: 0.8131
• F1: 0.8206

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=(0.9,0.99) and epsilon=1e-09
• 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
1.7127	1.0	321	0.9579	0.6987	0.6581	0.6987	0.6331
1.1596	2.0	642	0.7026	0.7299	0.7442	0.7299	0.7299
1.0337	3.0	963	0.6560	0.7549	0.7647	0.7549	0.7356
0.9695	4.0	1284	0.5708	0.7656	0.7963	0.7656	0.7758
0.9137	5.0	1605	0.6460	0.7611	0.7828	0.7611	0.7625
1.0053	6.0	1926	0.6020	0.7673	0.8049	0.7673	0.7797
0.896	7.0	2247	0.7087	0.7271	0.8055	0.7271	0.7477
0.7646	8.0	2568	0.6615	0.7441	0.8134	0.7441	0.7622
0.7262	9.0	2889	0.5611	0.7975	0.8190	0.7975	0.7985
0.7025	10.0	3210	0.5338	0.7975	0.8278	0.7975	0.8058
0.6138	11.0	3531	0.5143	0.8131	0.8359	0.8131	0.8206
0.5582	12.0	3852	0.6157	0.7864	0.8253	0.7864	0.7992
0.4736	13.0	4173	0.6899	0.8117	0.8253	0.8117	0.8007
0.4581	14.0	4494	0.6062	0.8128	0.8447	0.8128	0.8199
0.407	15.0	4815	0.5317	0.8308	0.8410	0.8308	0.8339
0.369	16.0	5136	0.6475	0.8197	0.8414	0.8197	0.8270
0.2855	17.0	5457	0.5153	0.8617	0.8572	0.8617	0.8578
0.2545	18.0	5778	0.5455	0.8436	0.8555	0.8436	0.8473
0.2221	19.0	6099	0.5955	0.8471	0.8624	0.8471	0.8516
0.2143	20.0	6420	0.5772	0.8575	0.8604	0.8575	0.8582
0.1619	21.0	6741	0.6021	0.8627	0.8567	0.8627	0.8572

Framework versions

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