TensorRT on L4: 40% Latency Reduction with Simple MLP

Minimal demo: Optimize PyTorch MLP for L4 GPUs via TensorRT.

Quick Start

1. git clone <repo> && cd tensorrt-l4-demo
2. pip install -r requirements.txt
3. python models/export_onnx.py
4. python tensorrt/build_engine.py
5. python benchmarks/benchmark.py → 40%+ drop!

Performance (L4, Batch=32)

Setup	Latency (ms)	Reduction
PyTorch	0.40-0.60	Baseline
TRT FP16	0.20-0.30	40-50%

L4 Tips: Use g2-standard-4; monitor with nvidia-smi.