EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[chore] Remove unused ep_moe cuda kernels (#9956)

Browse Source
This commit is contained in:
hlu1
2025-09-06 01:35:50 -07:00
committed by GitHub
parent 039cef76aa
commit 5f1eb20484
13 changed files with 4 additions and 1110 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
sgl-kernel/include/sgl_kernel_ops.h
View File

@@ -325,35 +325,6 @@ void prepare_moe_input(
const int64_t n,
const int64_t k);
void ep_moe_pre_reorder(
torch::Tensor input,
torch::Tensor gateup_input,
torch::Tensor src2dst,
torch::Tensor topk_ids,
torch::Tensor a1_scales,
int64_t start_expert_id,
int64_t end_expert_id,
int64_t topk,
bool use_per_token_if_dynamic);
void ep_moe_silu_and_mul(
torch::Tensor gateup_output,
torch::Tensor down_input,
torch::Tensor reorder_topk_ids,
torch::Tensor scales,
int64_t start_expert_id,
int64_t end_expert_id);
void ep_moe_post_reorder(
torch::Tensor down_output,
torch::Tensor output,
torch::Tensor src2dst,
torch::Tensor topk_ids,
torch::Tensor topk_weights,
int64_t start_expert_id,
int64_t end_expert_id,
int64_t topk);
void shuffle_rows(const torch::Tensor& input_tensor, const torch::Tensor& dst2src_map, torch::Tensor& output_tensor);
void apply_shuffle_mul_sum(