Assignment 1: Performance & Hardware Analysis
A deep dive into ResNet architectures (18, 32, 50) and SVM benchmarks. Features full analysis of FLOPs and hardware speedup factors.
DLOps Focus
Analysis of model depth vs. efficiency, FLOPs tracking, and hardware-aware training (CPU vs. GPU).
MLOps Focus
Automating model lifecycles, reproducibility, version control for data/models, and scalable deployment.
Current Assignments
Completed
Upcoming
Assignment 2: Containerization & CI/CD
Focusing on Dockerizing ML models and setting up automated GitHub Action pipelines for model testing.
Assignment 1 Live Metrics
Hardware Performance (FashionMNIST)
| Backend | Model | Accuracy | Time (ms) | FLOPs |
|---|---|---|---|---|
| GPU (CUDA) | ResNet-18 | 85.59% | 59,971 | 5.51e+08 |
| CPU (Intel) | ResNet-18 | 85.24% | 2,375,155 | 5.51e+08 |
Note: The GPU achieved a speedup of ~40x compared to the CPU while maintaining identical FLOPs.
Architecture Complexity
| Architecture | Params | GFLOPs |
|---|---|---|
| ResNet-18 | 11.2 M | 0.55 |
| ResNet-32 | 21.3 M | 1.14 |
| ResNet-50 | 23.5 M | 1.28 |