vit-dropout-0.2

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: 0.4888
• Accuracy: 0.8318
• Precision: 0.8450
• Recall: 0.8318
• F1: 0.8340

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: 1219
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
1.6129	1.0	321	0.8251	0.7223	0.7708	0.7223	0.7354
0.9753	2.0	642	0.7419	0.7115	0.7830	0.7115	0.7320
0.9289	3.0	963	0.5421	0.7968	0.7942	0.7968	0.7901
0.8978	4.0	1284	0.6563	0.7410	0.8112	0.7410	0.7569
0.8271	5.0	1605	0.5586	0.8058	0.8090	0.8058	0.8050
0.7757	6.0	1926	0.5340	0.8065	0.8220	0.8065	0.8113
0.6981	7.0	2247	0.5943	0.7791	0.8207	0.7791	0.7911
0.5954	8.0	2568	0.5836	0.7660	0.8270	0.7660	0.7824
0.5804	9.0	2889	0.4888	0.8318	0.8450	0.8318	0.8340
0.4894	10.0	3210	0.5404	0.8037	0.8459	0.8037	0.8145
0.4924	11.0	3531	0.4997	0.8166	0.8530	0.8166	0.8257
0.38	12.0	3852	0.5669	0.8356	0.8487	0.8356	0.8404
0.3109	13.0	4173	0.5983	0.8467	0.8492	0.8467	0.8415
0.3445	14.0	4494	0.5060	0.8620	0.8643	0.8620	0.8622
0.2457	15.0	4815	0.5823	0.8547	0.8564	0.8547	0.8520
0.2297	16.0	5136	0.5347	0.8620	0.8647	0.8620	0.8628
0.1701	17.0	5457	0.5561	0.8689	0.8665	0.8689	0.8656
0.186	18.0	5778	0.5319	0.8686	0.8688	0.8686	0.8679
0.1258	19.0	6099	0.6001	0.8655	0.8655	0.8655	0.8643

Framework versions

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