dinov2-large-2024_01_15-with_data_aug_batch-size32_epochs20_freeze

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

• Loss: 0.0891
• F1 Micro: 0.8422
• F1 Macro: 0.7067
• Roc Auc: 0.8958
• Accuracy: 0.5445
• Learning Rate: 0.0001

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	274	0.4728	0.5944	0.7681	0.1364	0.8536	0.001
0.2417	2.0	548	0.5028	0.6565	0.8040	0.1118	0.8701	0.001
0.2417	3.0	822	0.5122	0.6697	0.8124	0.1061	0.8763	0.001
0.1314	4.0	1096	0.5003	0.6643	0.8168	0.1062	0.8832	0.001
0.1314	5.0	1370	0.5178	0.6736	0.8176	0.1032	0.8783	0.001
0.1235	6.0	1644	0.5335	0.6928	0.8255	0.1027	0.8906	0.001
0.1235	7.0	1918	0.5237	0.6767	0.8205	0.1027	0.8774	0.001
0.1196	8.0	2192	0.5181	0.6758	0.8176	0.1027	0.8775	0.001
0.1196	9.0	2466	0.5335	0.6807	0.8224	0.0994	0.8765	0.001
0.117	10.0	2740	0.5167	0.6870	0.8283	0.1007	0.8937	0.001
0.1163	11.0	3014	0.5195	0.6925	0.8298	0.0971	0.8898	0.001
0.1163	12.0	3288	0.5230	0.7006	0.8282	0.0987	0.8861	0.001
0.1156	13.0	3562	0.5342	0.7065	0.8275	0.1017	0.8903	0.001
0.1156	14.0	3836	0.5276	0.6968	0.8243	0.1224	0.8851	0.001
0.1137	15.0	4110	0.5300	0.6958	0.8295	0.0981	0.8904	0.001
0.1137	16.0	4384	0.5398	0.7179	0.8412	0.0919	0.8981	0.0001
0.1091	17.0	4658	0.5433	0.7235	0.8431	0.0924	0.8997	0.0001
0.1091	18.0	4932	0.5447	0.7166	0.8448	0.0904	0.8984	0.0001
0.1026	19.0	5206	0.0903	0.8448	0.7248	0.8963	0.5443	0.0001
0.1026	20.0	5480	0.0887	0.8439	0.7176	0.8971	0.5422	0.0001

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.0+cu118
• Datasets 2.14.5
• Tokenizers 0.15.0