Pruna AI
summary
Pruna AI applies the open source pruna tooling to compress models by combining quantization, caching, compilation and factorization, turning a Flux Context image-edit model from 14 s on an H100 to 1.8 s (≈8× speedup) and cutting a 100k‑calls/day customer from ~280 kWh to ~35 kWh. At an inference‑provider scale optimizations yield 2–8× speedups (≈5× average), saving about $140k energy costs and roughly 90 years of GPU compute time per year.
AI-written summary · about
related talks
30mGreen AI in Cloud Native Ecosystems: Strategies for Sustainability and Efficiency
48mThe Power of Edge AI: Revolutionizing Future of Real-Time, Autonomous Processing
20mFireworks AI: Go from Hype to High - Value AI
01h 16mOptimizing workloads with Transformers and Diffusers
30mCompactifAI: Quantum-Inspired AI Model Compression
32mOptimizing AI/ML workloads for sustainability
30mCloud? No Thanks! I’m Gonna Run GenAI on My AI PC
20mProductionizing On-device AI Models