MobileNetV2-KD-VGGFace

This model is trained via KD from ViT on first 50k samples of VGGFace dataset. It achieves the following results on the evaluation set:

• Loss: 0.4919
• Accuracy: 0.7836

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: 5e-05
• train_batch_size: 24
• eval_batch_size: 24
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
2.7506	1.0	1667	2.2449	0.0726
2.105	2.0	3334	1.6904	0.2493
1.6544	3.0	5001	1.3206	0.4043
1.3357	4.0	6668	1.0675	0.5078
1.1104	5.0	8335	0.9302	0.5582
0.9287	6.0	10002	0.8738	0.5972
0.7899	7.0	11669	0.7972	0.6388
0.6738	8.0	13336	0.7074	0.6822
0.5803	9.0	15003	0.6630	0.7009
0.5038	10.0	16670	0.5855	0.735
0.4366	11.0	18337	0.5761	0.7415
0.3762	12.0	20004	0.5642	0.7496
0.3321	13.0	21671	0.5373	0.7652
0.2916	14.0	23338	0.5314	0.7625
0.2615	15.0	25005	0.6206	0.7281
0.2357	16.0	26672	0.5437	0.763
0.2153	17.0	28339	0.5335	0.763
0.1986	18.0	30006	0.4892	0.7869
0.1866	19.0	31673	0.5368	0.7645
0.1765	20.0	33340	0.4919	0.7836

Framework versions

• Transformers 4.42.4
• Pytorch 2.3.1+rocm6.0
• Datasets 2.20.0
• Tokenizers 0.19.1