vit-base-patch16-224-in21k-finetuned_v2024-7-24-frost

This model is a fine-tuned version of google/vit-base-patch16-224-in21k on the webdataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.0965
• Accuracy: 0.9637
• F1: 0.9118
• Precision: 0.9060
• Recall: 0.9177

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: 0.0002
• 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: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.0728	1.5625	100	0.0659	0.9841	0.9607	0.9692	0.9524
0.0871	3.125	200	0.1244	0.9566	0.8942	0.8922	0.8961
0.0999	4.6875	300	0.1043	0.9637	0.9126	0.8992	0.9264
0.0743	6.25	400	0.1043	0.9611	0.9043	0.9083	0.9004
0.0655	7.8125	500	0.0965	0.9637	0.9118	0.9060	0.9177
0.0559	9.375	600	0.1038	0.9619	0.9087	0.8917	0.9264
0.0517	10.9375	700	0.0972	0.9584	0.8998	0.8866	0.9134
0.0407	12.5	800	0.1120	0.9637	0.9111	0.9130	0.9091
0.0513	14.0625	900	0.1093	0.9558	0.8894	0.9095	0.8701
0.0378	15.625	1000	0.1197	0.9549	0.8889	0.8947	0.8831
0.0487	17.1875	1100	0.0955	0.9646	0.9138	0.9099	0.9177
0.0272	18.75	1200	0.1088	0.9566	0.8928	0.9027	0.8831
0.0241	20.3125	1300	0.0979	0.9637	0.9114	0.9095	0.9134
0.0311	21.875	1400	0.1134	0.9655	0.9158	0.9138	0.9177
0.0303	23.4375	1500	0.1092	0.9628	0.9079	0.92	0.8961
0.0225	25.0	1600	0.1121	0.9628	0.9083	0.9163	0.9004
0.0292	26.5625	1700	0.1149	0.9619	0.9071	0.9052	0.9091
0.0261	28.125	1800	0.1107	0.9619	0.9079	0.8983	0.9177
0.0166	29.6875	1900	0.1110	0.9611	0.9052	0.9013	0.9091

Framework versions

• Transformers 4.42.4
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1