vit-wonders-classification

• Loss: 0.0291
• Accuracy: 0.9948
• F1: 0.9955

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

• learning_rate: 0.0002
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Framework versions

• Transformers 4.55.1
• Pytorch 2.6.0+cu124
• Datasets 4.0.0
• Tokenizers 0.21.4

Training Results

• Final Accuracy: 99.48%
• Final F1-Score: 99.55%
• Training Loss: 0.1427
• Validation Loss: 0.0291
• Training Time: 15:37 minutes
• Epochs: 3

Methodology

• Base Model: google/vit-base-patch16-224-in21k
• Fine-tuning Approach: Full model fine-tuning
• Batch Size: 8 per device
• Learning Rate: 2e-4
• Evaluation Strategy: Every 100 steps
• Early Stopping: Based on accuracy improvement
• Dataset: 12-class landmark classification (3,846 images)
• Data Splits: 70% train, 20% validation, 10% test