vc-bantai-vit-withoutAMBI-adunest-v3

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: 0.8889
• Accuracy: 0.8218

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: 0.0005
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 200
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	0.38	100	0.8208	0.7147
No log	0.76	200	0.8861	0.7595
No log	1.14	300	0.4306	0.7910
No log	1.52	400	0.5222	0.8245
0.3448	1.9	500	0.8621	0.7602
0.3448	2.28	600	0.2902	0.8801
0.3448	2.66	700	0.3687	0.8426
0.3448	3.04	800	0.3585	0.8694
0.3448	3.42	900	0.6546	0.7897
0.2183	3.8	1000	0.3881	0.8272
0.2183	4.18	1100	0.9650	0.7709
0.2183	4.56	1200	0.6444	0.7917
0.2183	4.94	1300	0.4685	0.8707
0.2183	5.32	1400	0.4972	0.8506
0.157	5.7	1500	0.4010	0.8513
0.157	6.08	1600	0.4629	0.8419
0.157	6.46	1700	0.4258	0.8714
0.157	6.84	1800	0.4383	0.8573
0.157	7.22	1900	0.5324	0.8493
0.113	7.6	2000	0.3212	0.8942
0.113	7.98	2100	0.8621	0.8326
0.113	8.37	2200	0.6050	0.8131
0.113	8.75	2300	0.7173	0.7991
0.113	9.13	2400	0.5313	0.8125
0.0921	9.51	2500	0.6584	0.8158
0.0921	9.89	2600	0.8727	0.7930
0.0921	10.27	2700	0.4222	0.8922
0.0921	10.65	2800	0.5811	0.8265
0.0921	11.03	2900	0.6175	0.8372
0.0701	11.41	3000	0.3914	0.8835
0.0701	11.79	3100	0.3364	0.8654
0.0701	12.17	3200	0.6223	0.8359
0.0701	12.55	3300	0.7830	0.8125
0.0701	12.93	3400	0.4356	0.8942
0.0552	13.31	3500	0.7553	0.8232
0.0552	13.69	3600	0.9107	0.8292
0.0552	14.07	3700	0.6108	0.8580
0.0552	14.45	3800	0.5732	0.8567
0.0552	14.83	3900	0.5087	0.8614
0.0482	15.21	4000	0.8889	0.8218

Framework versions

• Transformers 4.20.1
• Pytorch 1.12.0+cu113
• Datasets 2.3.2
• Tokenizers 0.12.1