minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

[CPU] Add Qwen3.5 model optimization for CPU (#19484)

Browse Source 
Co-authored-by: Zheng, Beilei <beilei.zheng@intel.com>
Co-authored-by: Ma Mingfei <mingfei.ma@intel.com>
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
This commit is contained in:
jianan-gu
2026-04-26 10:12:36 -07:00
committed by GitHub
co-authored by Zheng, Beilei Ma Mingfei Xinyuan Tong
parent 7d49564431
commit 10fd0faccd
20 changed files with 768 additions and 209 deletions
Download Patch File Download Diff File Expand all files Collapse all files
sgl-kernel/csrc/cpu/moe.cpp
+7 -5
View File
@@ -495,13 +495,15 @@ void fused_experts_kernel_impl(
const scalar_t* __restrict__ B0 = packed_w1 + expert_id * stride_e + nb_upper * BLOCK_N * stride_n;
const scalar_t* __restrict__ B1 = packed_w1 + expert_id * stride_e + nb_lower * BLOCK_N * stride_n;
// 1.a load A
const int32_t* A_ids = sorted_ids + mb * BLOCK_M;
int64_t m_size = offsets[mb + 1] - offsets[mb];
for (int64_t m = 0; m < m_size; ++m) {
int32_t index = A_ids[m] / topk;
copy_stub(A + m * K, input + index * K, K);
if (nb_offset == 0) {
// 1.a load A
const int32_t* A_ids = sorted_ids + mb * BLOCK_M;
for (int64_t m = 0; m < m_size; ++m) {
int32_t index = A_ids[m] / topk;
copy_stub(A + m * K, input + index * K, K);
}
}
if (use_brgemm) {