Module execution_policy 

Typed execution-policy and admission authority contracts.

Structs

ExecutorAdmissionLimits

Immutable admission ceilings compiled into one model executor.

ExecutorAdmissionSnapshot

Runtime-owned admission state exposed to product health and validators.

Enums

AttentionExecutionPolicy

Product policy for selecting an attention provider implementation.

ExecutionResourceAuthority

Declares the authoritative owner of request-lifetime accelerator resources.

Constants

CUDA_NATIVE_ADAPTIVE_V1_MAX_SEQUENCE_TOKENS

Current CUDA native-adaptive boundary between single- and multi-partition addressed decode. The compiled provider and effective-config report consume this same constant.