PhyAI writing
Blog
Long-form thinking on intelligence that has to act in the physical world.
2026 Oct Update
Configurable kernel dispatch with Humming quantization, data-parallel serving across GPUs and hosts, PI0.5 PPO rollouts, and support for MolmoAct2 and Qwen-Image 2.1.
Kernel dispatch and Humming quantization in PhyAI
How PhyAI selects kernels for each workload, and how Humming brings configurable low-precision inference to the same execution path.
Data-parallel inference in PhyAI
Run complete policy replicas on multiple GPUs, distribute independent requests, and extend the replica pool across hosts.
Introducing PhyAI: Real-Time Physical AI at the Edge, Scalable Rollouts in the Cloud
A new foundation for fast, efficient AI deployment across edge devices and cloud infrastructure.
PhyAI Day 0 Support for MiniCPM-RobotManip
How PhyAI brought MiniCPM-RobotManip to optimized inference on release day, reaching 36.77 Hz on NVIDIA H20 and 47.85 Hz on RTX 5090.