vit-weldclassify

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

• Loss: 1.0639
• Accuracy: 0.8174

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
• num_epochs: 18
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.8311	0.8130	100	0.9623	0.4886
0.6016	1.6260	200	0.5911	0.7215
0.2602	2.4390	300	1.0585	0.6393
0.1643	3.2520	400	0.9470	0.7169
0.3754	4.0650	500	0.6054	0.8082
0.1446	4.8780	600	0.6845	0.7854
0.138	5.6911	700	0.9011	0.7534
0.033	6.5041	800	0.8366	0.8128
0.0538	7.3171	900	0.9102	0.7854
0.0144	8.1301	1000	0.8510	0.8128
0.0459	8.9431	1100	0.8610	0.8219
0.0022	9.7561	1200	0.9398	0.8082
0.0019	10.5691	1300	0.8714	0.8356
0.0015	11.3821	1400	1.0001	0.8128
0.0013	12.1951	1500	0.9926	0.8219
0.0012	13.0081	1600	1.0175	0.8219
0.0011	13.8211	1700	1.0323	0.8219
0.001	14.6341	1800	1.0453	0.8174
0.0009	15.4472	1900	1.0518	0.8174
0.0009	16.2602	2000	1.0585	0.8174
0.0009	17.0732	2100	1.0623	0.8174
0.0009	17.8862	2200	1.0639	0.8174

Framework versions

• Transformers 4.41.1
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1