A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 8, 07:30 UTC
[Bugfix] Replace c10::optional with std::optional in topk kernel (#34467)
Signed-off-by: Pushpinder Singh <pushpindersingh135@gmail.com>
Pushpinder Singh committed
Feb 13, 2026 at 08:30 UTC
bcd65c1f6a25ab76be325fbc0766eb074519a4fc
1 file changed
+1
-1
csrc/topk.cu
+1
-1
@@ -349,7 +349,7 @@ void setup_kernel_smem_once() {
349
void large_context_topk(
350
const torch::Tensor& logits, torch::Tensor& indices,
351
const torch::Tensor& seq_lens,
352
- c10::optional<torch::Tensor> row_starts = c10::nullopt) {
352
+ std::optional<torch::Tensor> row_starts = std::nullopt) {
353
TORCH_CHECK(logits.is_cuda(), "logits must be a CUDA tensor");
354
TORCH_CHECK(indices.is_cuda(), "indices must be a CUDA tensor");
355
TORCH_CHECK(seq_lens.is_cuda(), "seq_lens must be a CUDA tensor");