title	ML Performance Benchmark
emoji	🚀
colorFrom	blue
colorTo	indigo
sdk	docker
app_port	7860
pinned	false

ML Performance Benchmarking Platform

Interactive platform for comparing PyTorch optimization techniques in real-time.

🚀 Live Demo

Try it here: Hugging Face Space

Features

• 5 Optimization Modes: Baseline PyTorch, INT8 Quantization, TorchScript, ONNX Runtime, Dynamic Batching
• Real-time Metrics: Latency (p95), throughput (tokens/sec), memory usage
• Visual Comparisons: Interactive charts comparing performance
• Live Benchmarking: Test with custom prompts

Tech Stack

Backend:

• Python, PyTorch, FastAPI, Transformers
• ONNX Runtime for optimized inference
• SQLite for metrics storage

Frontend:

• React, TypeScript, Vite
• Tailwind CSS, Recharts

Performance Results

| Mode | Latency | Throughput | Speedup | |------|---------|------------|---------||------|---------|-----tok/s |------|---------|------------8.|------|--2.|------|---line |------|------tok/s | 1.0x |

Local Development

Backend

cd backend
pip install -r requpip install -r requpip install -r requpip install -r requpip install -r requpip inpip install -r rv

Author

Built by Siddharth

License

MIT