Module graph_walker 

CpuBackend::execute_graph — walk a GraphProto body (the kind the compiler’s collapse_backend_subgraphs pass emits inside every BackendSubgraph_* FunctionProto) and run each NodeProto through the existing kernel dispatch.

No fancy scheduling — ONNX guarantees GraphProto.node is already topologically ordered, so a linear walk suffices.

Enums

BackendError

Failures execute_graph may surface alongside the kernel-level OpErrors already routed through the existing dispatch.

Functions

execute_graph

Run every NodeProto in graph.node in order, threading a HashMap<String, CpuTensor> value env. Returns the subset of env named in graph.output.