vit-class-weight

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: 0.4472
• Accuracy: 0.8478
• Precision: 0.8582
• Recall: 0.8478
• F1: 0.8483

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
1.5485	1.0	321	0.8743	0.6813	0.7810	0.6813	0.7087
0.9628	2.0	642	0.7893	0.6907	0.7945	0.6907	0.7178
0.8902	3.0	963	0.5577	0.7926	0.7956	0.7926	0.7835
0.8477	4.0	1284	0.5734	0.7611	0.8190	0.7611	0.7770
0.7773	5.0	1605	0.6590	0.7431	0.8052	0.7431	0.7590
0.6953	6.0	1926	0.5321	0.8100	0.8298	0.8100	0.8167
0.6122	7.0	2247	0.5331	0.8044	0.8280	0.8044	0.8093
0.5548	8.0	2568	0.6589	0.7649	0.8313	0.7649	0.7832
0.512	9.0	2889	0.4548	0.8395	0.8445	0.8395	0.8402
0.449	10.0	3210	0.4472	0.8478	0.8582	0.8478	0.8483
0.4012	11.0	3531	0.5304	0.8287	0.8509	0.8287	0.8353
0.3584	12.0	3852	0.5620	0.8454	0.8576	0.8454	0.8468
0.2829	13.0	4173	0.6837	0.8436	0.8490	0.8436	0.8359
0.2761	14.0	4494	0.6061	0.8509	0.8643	0.8509	0.8541
0.2192	15.0	4815	0.5223	0.8637	0.8662	0.8637	0.8639
0.1755	16.0	5136	0.5640	0.8558	0.8684	0.8558	0.8591
0.1568	17.0	5457	0.5585	0.8682	0.8736	0.8682	0.8695
0.1674	18.0	5778	0.5645	0.8724	0.8735	0.8724	0.8707
0.1022	19.0	6099	0.5931	0.8745	0.8740	0.8745	0.8737
0.1487	20.0	6420	0.6107	0.8717	0.8736	0.8717	0.8722

Framework versions

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