adapters-vs-lora-cifar10

This repository provides a modular, reproducible, and enterprise-grade framework for benchmarking Adapter and LoRA architectures against a ViT-Tiny baseline on the CIFAR-10 dataset. Designed for research and production, it features robust logging, error handling, and statistical reporting.

Features

• Modular codebase with clear separation of concerns
• Baseline, Adapter, and LoRA model variants
• Sophisticated logging and exception handling
• Reproducible experiments (seed management, config)
• Comprehensive statistics and visualization utilities
• Extensible for new models, datasets, or experiment types
• Ready for CI/CD and large-scale benchmarking

Project Structure

adapters_vs_lora_cifar10/
  config.py        # Experiment configuration and seed management
  data.py          # Data loading, transformation, and validation
  visualize.py     # Visualization utilities
  model/
    model_utils.py # Model architectures (ViT, Adapters, LoRA integration)
    train_utils.py # Training and evaluation logic
  logger.py        # Logging and error handling
  stats.py         # Statistics and metrics
scripts/
  run_experiment.py # Entry point for running all experiments
tests/             # Unit and integration tests
requirements.txt   # Python dependencies
README.md          # Project documentation

Getting Started

1. Clone the repository
2. Create and activate a virtual environment:
   • Windows: python -m venv .venv && .venv\Scripts\activate
   • Linux/Mac: python3 -m venv .venv && source .venv/bin/activate
3. Install dependencies: pip install -r requirements.txt
4. Run all experiments (from project root):
   • python -m scripts.run_experiment

Experiment Variants

• Baseline: ViT-Tiny + classification head
• Adapter: ViT-Tiny with integrated Adapter layers (train only adapters and head)
• LoRA: ViT-Tiny with LoRA applied to QKV projections (requires peft)

Results are saved in results/all_results.json.

Extending the Framework

• Add new model variants in adapters_vs_lora_cifar10/model/model_utils.py
• Add new training/evaluation logic in train_utils.py
• Add new datasets in data.py
• Add new experiment scripts in scripts/

Troubleshooting

• Always run scripts as modules from the project root (e.g., python -m scripts.run_experiment)
• If you see ModuleNotFoundError, check your working directory and PYTHONPATH
• For LoRA experiments, ensure peft is installed (pip install peft)

License

MIT