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 16, 20:45 UTC
[Bugfix][Kernel] Promote another index to int64_t (#6838)
Tyler Michael Smith committed
Jul 26, 2024 at 14:41 UTC
50704f52c4643777fb0e5dc99f6c048dd9f54f2d
1 file changed
+1
-1
csrc/quantization/fp8/common.cu
+1
-1
@@ -48,7 +48,7 @@ __global__ void segmented_max_reduction(float* __restrict__ scale,
48
const scalar_t* __restrict__ input,
49
int64_t num_elems) {
50
__shared__ float cache[1024];
51
- int i = blockDim.x * blockIdx.x + threadIdx.x;
51
+ int64_t i = blockDim.x * blockIdx.x + threadIdx.x;
52
53
// First store maximum for all values processes by
54
// the current thread in cache[threadIdx.x]