biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Apply fixes for CUDA 13 (#24599)

Browse Source 
Signed-off-by: Aidyn-A <aidyn.b.aitzhan@gmail.com>
This commit is contained in:
Aidyn-A
2025-09-17 17:15:42 +04:00
committed by GitHub
parent 9fccd04e30
commit bfe9380161
8 changed files with 47 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
csrc/quantization/fp8/common.cu
View File

@@ -1,15 +1,10 @@
#include "common.cuh"
#include "dispatch_utils.h"
#include "../../cub_helpers.h"
#include "../vectorization_utils.cuh"
#include <c10/cuda/CUDAGuard.h>
#include <ATen/cuda/Exceptions.h>
#ifndef USE_ROCM
#include <cub/cub.cuh>
#else
#include <hipcub/hipcub.hpp>
#endif
namespace vllm {
template <typename scalar_t, typename fp8_type>
@@ -116,7 +111,7 @@ __global__ void dynamic_per_token_scaled_fp8_quant_kernel_strided(
using BlockReduce = cub::BlockReduce<float, 256>;
__shared__ typename BlockReduce::TempStorage tmp;
const float block_max =
BlockReduce(tmp).Reduce(absmax_val, cub::Max{}, blockDim.x);
BlockReduce(tmp).Reduce(absmax_val, CubMaxOp{}, blockDim.x);
__shared__ float token_scale;
if (tid == 0) {