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 14, 05:30 UTC
[MISC] replace c10::optional with std::optional (#25602)
Signed-off-by: Shiyan Deng <dsy842974287@meta.com>
Shiyan Deng committed
Sep 24, 2025 at 16:56 UTC
5c1e496a75138f6b7663cf56d733b6f9e06d7d66
2 files changed
+4
-4
csrc/rocm/ops.h
+2
-2
@@ -6,11 +6,11 @@ torch::Tensor LLMM1(at::Tensor& in_a, at::Tensor& in_b,
6
const int64_t rows_per_block);
7
8
torch::Tensor wvSplitK(const at::Tensor& in_a, const at::Tensor& in_b,
9
- const c10::optional<at::Tensor>& in_bias,
9
+ const std::optional<at::Tensor>& in_bias,
10
const int64_t CuCount);
11
12
void wvSplitKQ(const at::Tensor& in_a, const at::Tensor& in_b,
13
- const c10::optional<at::Tensor>& in_bias, at::Tensor& out_c,
13
+ const std::optional<at::Tensor>& in_bias, at::Tensor& out_c,
14
const at::Tensor& scale_a, const at::Tensor& scale_b,
15
const int64_t CuCount);
16
csrc/rocm/skinny_gemms.cu
+2
-2
@@ -1271,7 +1271,7 @@ int mindiv(int N, int div1, int div2) {
1271
}
1272
1273
torch::Tensor wvSplitK(const at::Tensor& in_a, const at::Tensor& in_b,
1274
- const c10::optional<at::Tensor>& in_bias,
1274
+ const std::optional<at::Tensor>& in_bias,
1275
const int64_t CuCount) {
1276
auto M_in = in_a.size(0);
1277
auto K_in = in_a.size(1);
@@ -1729,7 +1729,7 @@ __global__ void wvSplitKQ_hf_(const int K, const int Kp, const int M,
1729
#endif // defined(__HIP__MI3XX__) TODO: Add NAVI support
1730
1731
void wvSplitKQ(const at::Tensor& in_a, const at::Tensor& in_b,
1732
- const c10::optional<at::Tensor>& in_bias, at::Tensor& out_c,
1732
+ const std::optional<at::Tensor>& in_bias, at::Tensor& out_c,
1733
const at::Tensor& scale_a, const at::Tensor& scale_b,
1734
const int64_t CuCount) {
1735
static c10::ScalarType kFp8Type = is_fp8_ocp()