PyTorch ML Graph Optimizer

Optimize PyTorch models with compiler-level techniques.

This project demonstrates graph-level optimizations for deep learning models using PyTorch. It fuses convolution + batch normalization (+ optional ReLU), removes redundant layers, profiles latency, and generates visual summaries of the optimization effect.

---

Features

• Conv+BN(+ReLU) fusion for faster inference
• Remove nn.Identity layers for cleaner models
• Latency profiling on CPU or GPU
• Visual output:
  • Layer composition before vs after optimization
  • Inference latency comparison
• TorchScript export for compiler-ready models