Accelerated Machine Learning with Google Cloud and NVIDIA GPUs
Link to open source: https://github.com/samirlevate8788-creator/accelerated-ml-with-gpus/archive/refs/tags/v1.0.0.zip
Link to Live Project: https://samirlevate8788-creator.github.io/accelerated-ml-with-gpus/
Complete implementation of the official Google Developers Codelab:
"Accelerated Machine Learning with Google Cloud and NVIDIA."
Built an end-to-end GPU-accelerated regression pipeline on 2.96M+ NYC Yellow
Taxi trip records to predict tip amounts using:
• NVIDIA cuDF (zero-code-change pandas GPU acceleration)
• NVIDIA cuML (zero-code-change scikit-learn GPU acceleration)
• XGBoost with CUDA native GPU (device='cuda', tree_method='hist')
• 3-fold cross-validated ensemble (XGBoost + Random Forest + Linear Regression)
Key Results:
- Ensemble RMSE: $0.9084 (Lift: +$0.0131 over best single model)
- End-to-End Speedup: 13.1x faster on NVIDIA L4 GPU vs n2-standard CPU
- Random Forest: 18.4x speedup
- XGBoost: 12.8x speedup
- Feature Engineering: 7.1x speedup
Includes: Interactive live web dashboard, CLI pipeline scripts,
CPU vs GPU benchmark visualizer, and GitHub Actions CI/CD.
