ML Performance Profiler

A lightweight PyTorch benchmarking and profiling toolkit for analyzing neural network performance.

Features

• PyTorch Profiler integration
• CPU execution profiling
• Chrome trace export
• Memory utilization monitoring
• Throughput benchmarking
• CSV report generation
• Performance visualization

Project Structure

ML-Performance-Profiler/
├── benchmark.py
├── metrics.py
├── throughput.py
├── report.py
├── visualize.py
├── reports/
│   ├── results.csv
│   └── throughput.png
├── traces/
│   └── trace.json
└── requirements.txt

Installation

pip install -r requirements.txt

Run Benchmark

python benchmark.py

Example Output

Elapsed Time: 8.32s
Throughput: 3077.28 samples/sec
Memory Usage: 368.97 MB
Trace saved to traces/trace.json

Generate Visualization

python visualize.py

Output:

reports/
├── results.csv
└── throughput.png

Metrics Collected

Metric	Description
CPU Time	Operator-level execution time
Memory Usage	Process memory consumption
Throughput	Samples processed per second
Execution Trace	Chrome trace for performance analysis

Technologies

• Python
• PyTorch
• Pandas
• Matplotlib
• psutil

Future Improvements

• GPU profiling support
• Multi-model benchmarking
• Interactive dashboard
• Distributed training analysis
• Automated optimization suggestions

Benchmark Results

Execution Trace Analysis

PyTorch execution trace visualized using Perfetto.