use abosolute path

Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-08-27 21:44:27 +00:00
parent e92676ef4e
commit afe23a2990
3 changed files with 7 additions and 7 deletions
--- a/csrc/quantization/w8a8/fp8/per_token_group_quant.cu
+++ b/csrc/quantization/w8a8/fp8/per_token_group_quant.cu
@@ -1,6 +1,6 @@
 #include <ATen/cuda/CUDAContext.h>

-#include "../per_token_group_quant_8bit.h"
+#include "quantization/w8a8/per_token_group_quant_8bit.h"

 #include <cmath>

@@ -8,9 +8,9 @@

 #include <torch/all.h>

-#include "../../vectorization.cuh"
-#include "../../vectorization_utils.cuh"
-#include "../../../dispatch_utils.h"
+#include "quantization/vectorization.cuh"
+#include "quantization/vectorization_utils.cuh"
+#include "dispatch_utils.h"

 __device__ __forceinline__ float GroupReduceMax(float val, const int tid) {
  unsigned mask = 0xffff;