Module executor 

Graph executor — runs a fused IR graph on CPU using the arena + kernels.

The executor is the runtime hot path. For a 6-layer BERT, it makes ~24 kernel calls total (one per fused node). Everything else is inside the kernels — SIMD, BLAS, pre-allocated arena buffers.

Structs

ExternalBuffers

External data provided at runtime (model weights + inputs).

Functions

execute

Execute a compiled graph on CPU.