mgbam/ASTMODEL

resnet18 (AST-Trained)

Trained with 65% less energy than standard training ⚡

Model Details

• Architecture: resnet18
• Dataset: CIFAR-10
• Training Method: Adaptive Sparse Training (AST)
• Target Activation Rate: 35%

Performance

• Accuracy: 6809.00%
• Energy Savings: 65%
• Training Epochs: 10

Sustainability Report

This model was trained using Adaptive Sparse Training, which dynamically selects the most important training samples. This resulted in:

• ⚡ 65% energy savings compared to standard training
• 🌍 Lower carbon footprint
• ⏱️ Faster training time
• 🎯 Maintained accuracy (minimal degradation)

How to Use

import torch
from torchvision import models

# Load model
model = models.resnet18(num_classes=10)
model.load_state_dict(torch.load("pytorch_model.bin"))
model.eval()

# Inference
# ... (your inference code)

Training Details

AST Configuration:

• Target Activation Rate: 35%
• Adaptive PI Controller: Enabled
• Mixed Precision (AMP): Enabled

Reproducing This Model

pip install adaptive-sparse-training

python -c "
from adaptive_sparse_training import AdaptiveSparseTrainer, ASTConfig
config = ASTConfig(target_activation_rate=0.35)
# ... (full training code)
"

Citation

If you use this model or AST, please cite:

@software{adaptive_sparse_training,
    title={Adaptive Sparse Training},
    author={Idiakhoa, Oluwafemi},
    year={2024},
    url={https://github.com/oluwafemidiakhoa/adaptive-sparse-training}
}

Acknowledgments

Trained using the adaptive-sparse-training package. Special thanks to the PyTorch and HuggingFace communities.

---

This model card was auto-generated by the AST Training Dashboard.