dinov2-large-2024_01_14-without_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: 0.0904
• F1 Micro: 0.8447
• F1 Macro: 0.7156
• Roc Auc: 0.9011
• Accuracy: 0.5459
• 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	0.4696	0.5445	0.7663	0.1265	0.8401	0.001
0.2337	2.0	548	0.5115	0.6504	0.8026	0.1108	0.8694	0.001
0.2337	3.0	822	0.5178	0.6963	0.8184	0.1064	0.8804	0.001
0.1259	4.0	1096	0.5188	0.6838	0.8164	0.1062	0.8808	0.001
0.1259	5.0	1370	0.4965	0.6749	0.8157	0.1069	0.8849	0.001
0.1181	6.0	1644	0.5213	0.6806	0.8223	0.1028	0.8816	0.001
0.1181	7.0	1918	0.5269	0.6718	0.8253	0.0992	0.8857	0.001
0.1146	8.0	2192	0.5216	0.6811	0.8224	0.1004	0.8815	0.001
0.1146	9.0	2466	0.5230	0.6845	0.8302	0.1019	0.8923	0.001
0.1123	10.0	2740	0.5279	0.6775	0.8181	0.1021	0.8754	0.001
0.11	11.0	3014	0.5429	0.6897	0.8290	0.0960	0.8815	0.001
0.11	12.0	3288	0.5373	0.6825	0.8316	0.0967	0.8896	0.001
0.1098	13.0	3562	0.5328	0.6961	0.8254	0.1009	0.8838	0.001
0.1098	14.0	3836	0.0992	0.8278	0.7092	0.8834	0.5331	0.001
0.1166	15.0	4110	0.0991	0.8230	0.6923	0.8796	0.5335	0.001
0.1166	16.0	4384	0.0918	0.8395	0.7124	0.8932	0.5391	0.0001
0.1072	17.0	4658	0.0907	0.8439	0.7179	0.8990	0.5471	0.0001
0.1072	18.0	4932	0.0891	0.8441	0.7237	0.8971	0.5447	0.0001
0.1017	19.0	5206	0.0892	0.8479	0.7277	0.9027	0.5478	0.0001
0.1017	20.0	5480	0.0883	0.8484	0.7310	0.9043	0.5509	0.0001

Framework versions

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