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
Revert "[Refactor] Fix Compile Warning #1444-D (#21208)" (#21384)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Jul 22, 2025 at 11:22 UTC
226b452a2059166a00d4d3f620288fced4720c55
1 file changed
+1
-2
csrc/moe/topk_softmax_kernels.cu
+1
-2
@@ -20,7 +20,6 @@
20
#include <ATen/cuda/CUDAContext.h>
21
#include <c10/cuda/CUDAGuard.h>
22
#include "../cuda_compat.h"
23
-#include <cuda/std/functional>
23
24
#ifndef USE_ROCM
25
#include <cub/util_type.cuh>
@@ -63,7 +62,7 @@ __launch_bounds__(TPB) __global__
62
63
const int thread_row_offset = blockIdx.x * num_cols;
64
66
- cuda::std::plus<float> sum;
65
+ cub::Sum sum;
66
float threadData(-FLT_MAX);
67
68
// Don't touch finished rows.