Runtime Benchmark Options

struct RuntimeBenchmarkOptions

Optional experiment controls. Defaults preserve normal request semantics.

Public Functions

inline bool enabled() const noexcept

Public Members

int32_t maxDecodeRounds = {0}

Zero leaves the request token/EOS budget in control.

int32_t forcedAcceptanceLength = {-1}

Total accepted tokens including the bonus; -1 uses natural acceptance.

bool gpuTiming = {true}

CUDA-event stage timing while profiling; acceptance records do not depend on it.