swin-tiny-finetuned-cifar100

This model is a fine-tuned version of microsoft/swin-tiny-patch4-window7-224 on the cifar100 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4223
• Accuracy: 0.8735

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: 4e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20 (with early stopping)

Training results

Training Loss	Epoch	Step	Accuracy	Validation Loss
0.6439	1.0	781	0.8138	0.6126
0.6222	2.0	1562	0.8393	0.5094
0.2912	3.0	2343	0.861	0.4452
0.2234	4.0	3124	0.8679	0.4330
0.121	5.0	3905	0.8735	0.4223
0.2589	6.0	4686	0.8622	0.4775
0.1419	7.0	5467	0.8642	0.4900
0.1513	8.0	6248	0.8667	0.4956

Framework versions

• Transformers 4.20.1
• Pytorch 1.11.0
• Datasets 2.1.0
• Tokenizers 0.12.1