dinov2-giant-2024_01_02-kornia_img-size518_batch-size32_epochs20_freeze

This model is a fine-tuned version of facebook/dinov2-giant on the multilabel_complete_dataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.1266
• F1 Micro: 0.8142
• F1 Macro: 0.7719
• Roc Auc: 0.8801
• Accuracy: 0.5121
• Learning Rate: 0.001

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: 0.01
• 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: 20

Training results

Training Loss	Epoch	Step	Accuracy	F1 Macro	F1 Micro	Validation Loss	Roc Auc	Rate
No log	1.0	268	0.4055	0.5114	0.6258	0.2231	0.7463	0.01
0.2273	2.0	536	0.3812	0.4511	0.6106	0.2505	0.7360	0.01
0.2273	3.0	804	0.4176	0.6952	0.7531	0.1782	0.8425	0.01
0.196	4.0	1072	0.4241	0.6667	0.7646	0.1578	0.8562	0.01
0.196	5.0	1340	0.3551	0.6463	0.7290	0.1978	0.8616	0.01
0.1916	6.0	1608	0.4548	0.6155	0.7534	0.1570	0.8332	0.01
0.1916	7.0	1876	0.4076	0.7034	0.7711	0.1704	0.8893	0.01
0.1935	8.0	2144	0.4487	0.7240	0.7783	0.1584	0.8759	0.01
0.1935	9.0	2412	0.4434	0.7026	0.7725	0.1614	0.8787	0.01
0.1945	10.0	2680	0.4366	0.6245	0.7438	0.1569	0.8239	0.01
0.1945	11.0	2948	0.4298	0.6986	0.7639	0.1666	0.8614	0.01
0.1951	12.0	3216	0.4477	0.6291	0.7448	0.1585	0.8242	0.01
0.1951	13.0	3484	0.1565	0.7624	0.6650	0.8443	0.4380	0.01
0.1953	14.0	3752	0.1728	0.6728	0.5022	0.7639	0.4466	0.01
0.1945	15.0	4020	0.1565	0.7441	0.6524	0.8177	0.4555	0.01
0.1945	16.0	4288	0.1576	0.7515	0.6439	0.8311	0.4580	0.01
0.1929	17.0	4556	0.1701	0.7359	0.5707	0.8337	0.4312	0.01
0.1929	18.0	4824	0.1599	0.7531	0.6534	0.8451	0.4230	0.01
0.1952	19.0	5092	0.1603	0.7347	0.6658	0.8118	0.4548	0.01
0.1952	20.0	5360	0.1276	0.8134	0.7677	0.8759	0.5263	0.001

Framework versions

• Transformers 4.34.1
• Pytorch 2.1.0+cu118
• Datasets 2.14.5
• Tokenizers 0.14.1