PTO Runtime
Task-graph runtime for distributed tensor compilation and coordinated host-device execution
PTO Runtime focuses on the execution side of compiled tensor workloads. It turns dependency-rich task graphs into coordinated runtime execution across host control, device-side scheduling, and compute kernels.
compiled task graph
|
v
runtime coordination
|
v
scalable device execution
The project aims to make large tensor programs feel like organized flows of work instead of disconnected kernels. By treating execution as a graph problem, PTO Runtime creates a path from compilation output to structured, scalable system behavior.