Greedy Argmax Kernel#

struct ArgmaxPair#

Public Members

float value = {-std::numeric_limits<float>::infinity()}#
int32_t index = {-1}#
int32_t valid = {0}#
cudaError_t trt_edgellm::safety::launchGreedyArgmax(
float const *logits,
int32_t vocabularySize,
ArgmaxPair *result,
cudaStream_t stream
) noexcept#