minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

update pre-commit config (#18860)

Browse Source
This commit is contained in:
SoluMilken
2026-02-16 00:18:31 +08:00
committed by GitHub
parent f7603203b0
commit 07a24f1a38
135 changed files with 239 additions and 198 deletions
Download Patch File Download Diff File Expand all files Collapse all files
sgl-kernel/csrc/attention/merge_attn_states.cu
+4 -2
View File
@@ -176,8 +176,10 @@ void merge_attn_states_launcher(
LAUNCH_MERGE_ATTN_STATES(scalar_t, NUM_THREADS);
}
#define CALL_MERGE_ATTN_STATES_LAUNCHER(scalar_t) \
{ merge_attn_states_launcher<scalar_t>(v_a, s_a, v_b, s_b, v_merged, s_merged); }
#define CALL_MERGE_ATTN_STATES_LAUNCHER(scalar_t) \
{ \
merge_attn_states_launcher<scalar_t>(v_a, s_a, v_b, s_b, v_merged, s_merged); \
}
void merge_state_v2(
at::Tensor v_a, at::Tensor s_a, at::Tensor v_b, at::Tensor s_b, at::Tensor v_merged, at::Tensor s_merged) {
sgl-kernel/csrc/common_extension_rocm.cc
+2 -2
View File
@@ -65,9 +65,9 @@ TORCH_LIBRARY_EXPAND(sgl_kernel, m) {
m.impl("all_reduce_unreg", torch::kCUDA, &all_reduce_unreg);
// Deterministic all-reduce for ROCm
extern void deterministic_all_reduce_reg(int64_t _fa, torch::Tensor & inp, torch::Tensor & out);
extern void deterministic_all_reduce_reg(int64_t _fa, torch::Tensor& inp, torch::Tensor& out);
extern void deterministic_all_reduce_unreg(
int64_t _fa, torch::Tensor & inp, torch::Tensor & reg_buffer, torch::Tensor & out);
int64_t _fa, torch::Tensor& inp, torch::Tensor& reg_buffer, torch::Tensor& out);
m.def("deterministic_all_reduce_reg(int fa, Tensor inp, Tensor! out) -> ()");
m.impl("deterministic_all_reduce_reg", torch::kCUDA, &deterministic_all_reduce_reg);
sgl-kernel/csrc/gemm/per_token_group_quant_8bit_v2.cu
+1 -1
View File
@@ -452,7 +452,7 @@ void sgl_per_token_group_quant_8bit_v2(
#define LAUNCH_KERNEL(GROUP_SIZE, T, DST_DTYPE) \
do { \
constexpr int THREADS_PER_SUBWARP = GROUP_SIZE / 16; \
TORCH_CHECK(THREADS_PER_SUBWARP* INPUT_PRIMARY_VEC_NUM_BYTES == group_size * sizeof(T)); \
TORCH_CHECK(THREADS_PER_SUBWARP * INPUT_PRIMARY_VEC_NUM_BYTES == group_size * sizeof(T)); \
\
using dst_dtype_info = DtypeInfo<DST_DTYPE>; \
CHECK_EQ(dst_dtype_info::MIN, min_8bit); \