Dflash Decode Utils

int32_t trt_edgellm::rt::dflash_utils::runtimeBlockSize(

DeploymentConfig const &deployment,

)

Return the DFlash draft block horizon used by runtime draft forward.

DeploymentConfig resolves explicit/user-supplied DFlash horizons before the decoder is created, so the runtime reads one consolidated value.

bool trt_edgellm::rt::dflash_utils::shouldUseDDTree(

DeploymentConfig const &deployment,

)

DFlash uses a linear-tree proposal when draftingTopK == 1 and a branching DDTree proposal when draftingTopK > 1.