swin-large-patch4-window7-224-fv-finetuned-memes

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

• Loss: 0.6502
• Accuracy: 0.8601
• Precision: 0.8582
• Recall: 0.8601
• F1: 0.8583

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: 0.00012
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 256
• 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

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
1.2077	0.99	20	0.9499	0.6461	0.6764	0.6461	0.5863
0.5687	1.99	40	0.5365	0.7975	0.8018	0.7975	0.7924
0.3607	2.99	60	0.4007	0.8423	0.8419	0.8423	0.8398
0.203	3.99	80	0.3751	0.8509	0.8502	0.8509	0.8503
0.1728	4.99	100	0.4168	0.8509	0.8519	0.8509	0.8506
0.0963	5.99	120	0.4351	0.8586	0.8573	0.8586	0.8555
0.0956	6.99	140	0.4415	0.8547	0.8542	0.8547	0.8541
0.079	7.99	160	0.5312	0.8501	0.8475	0.8501	0.8459
0.0635	8.99	180	0.5376	0.8601	0.8578	0.8601	0.8577
0.0593	9.99	200	0.5060	0.8609	0.8615	0.8609	0.8604
0.0656	10.99	220	0.4997	0.8617	0.8573	0.8617	0.8587
0.0561	11.99	240	0.5430	0.8586	0.8604	0.8586	0.8589
0.0523	12.99	260	0.5354	0.8624	0.8643	0.8624	0.8626
0.0489	13.99	280	0.5539	0.8609	0.8572	0.8609	0.8577
0.0487	14.99	300	0.5785	0.8609	0.8591	0.8609	0.8591
0.0485	15.99	320	0.6186	0.8601	0.8578	0.8601	0.8573
0.0518	16.99	340	0.6342	0.8624	0.8612	0.8624	0.8606
0.0432	17.99	360	0.6302	0.8586	0.8598	0.8586	0.8580
0.0469	18.99	380	0.6323	0.8617	0.8606	0.8617	0.8604
0.0426	19.99	400	0.6502	0.8601	0.8582	0.8601	0.8583

Framework versions

• Transformers 4.24.0.dev0
• Pytorch 1.11.0+cu102
• Datasets 2.6.1.dev0
• Tokenizers 0.13.1