vit-base-patch16-384-finetuned_v2024-7-25-frost

This model is a fine-tuned version of google/vit-base-patch16-384 on the webdataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.0795
• Accuracy: 0.9747
• F1: 0.9373
• Precision: 0.9342
• Recall: 0.9404

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
0.0624	1.7544	100	0.0458	0.9867	0.9665	0.9774	0.9558
0.0729	3.5088	200	0.0942	0.9689	0.9220	0.9303	0.9139
0.0566	5.2632	300	0.0802	0.972	0.9311	0.9221	0.9404
0.051	7.0175	400	0.0965	0.9631	0.9066	0.9243	0.8896
0.0686	8.7719	500	0.0795	0.9747	0.9373	0.9342	0.9404
0.0271	10.5263	600	0.0935	0.9693	0.9239	0.9229	0.9249
0.0273	12.2807	700	0.0975	0.9716	0.9300	0.9219	0.9382
0.0445	14.0351	800	0.0910	0.9698	0.9248	0.9268	0.9227
0.0217	15.7895	900	0.0942	0.9698	0.9243	0.9326	0.9161
0.0257	17.5439	1000	0.0906	0.9684	0.9210	0.9283	0.9139
0.0188	19.2982	1100	0.1028	0.9676	0.9181	0.9338	0.9029
0.0196	21.0526	1200	0.1020	0.9698	0.9244	0.9306	0.9183
0.025	22.8070	1300	0.1005	0.9702	0.9258	0.9289	0.9227
0.009	24.5614	1400	0.0976	0.9729	0.9324	0.9356	0.9294
0.0184	26.3158	1500	0.0987	0.9716	0.9290	0.9332	0.9249
0.0048	28.0702	1600	0.0958	0.972	0.9301	0.9353	0.9249
0.0072	29.8246	1700	0.0948	0.972	0.9301	0.9353	0.9249

Framework versions

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