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 15, 01:00 UTC
[Refactor] Fix Compile Warning #1444-D (#21208)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Jul 22, 2025 at 02:33 UTC
6e5b5ca580e60a3b7a6c5613ae7d8676f2c8cbc6
1 file changed
+2
-1
csrc/moe/topk_softmax_kernels.cu
+2
-1
@@ -20,6 +20,7 @@
20
#include <ATen/cuda/CUDAContext.h>
21
#include <c10/cuda/CUDAGuard.h>
22
#include "../cuda_compat.h"
23
+#include <cuda/std/functional>
24
25
#ifndef USE_ROCM
26
#include <cub/util_type.cuh>
@@ -62,7 +63,7 @@ __launch_bounds__(TPB) __global__
63
64
const int thread_row_offset = blockIdx.x * num_cols;
65
65
- cub::Sum sum;
66
+ cuda::std::plus<float> sum;
67
float threadData(-FLT_MAX);
68
69
// Don't touch finished rows.