minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 14 Packages Projects Releases Wiki Activity

[kernel slimming] Clean many useless sgl-kernel deprecated kernels (#20277)

Browse Source
This commit is contained in:
Xiaoyu Zhang
2026-03-14 16:45:54 +08:00
committed by GitHub
parent 75a7879fd4
commit 25e38216b6
26 changed files with 60 additions and 1483 deletions
Download Patch File Download Diff File Expand all files Collapse all files
sgl-kernel/csrc/common_extension.cc
-12
View File
@@ -84,12 +84,6 @@ TORCH_LIBRARY_FRAGMENT(sgl_kernel, m) {
m.def("gelu_and_mul(Tensor! out, Tensor input) -> ()");
m.impl("gelu_and_mul", torch::kCUDA, &gelu_and_mul);
m.def(
"apply_rope_pos_ids_cos_sin_cache(Tensor q, Tensor k, Tensor! q_rope, Tensor! k_rope, Tensor cos_sin_cache, "
"Tensor pos_ids, bool interleave, bool enable_pdl, "
"Tensor? v, Tensor!? k_buffer, Tensor!? v_buffer, Tensor? kv_cache_loc) -> ()");
m.impl("apply_rope_pos_ids_cos_sin_cache", torch::kCUDA, &apply_rope_pos_ids_cos_sin_cache);
m.def(
"rotary_embedding(Tensor positions, Tensor! query,"
" Tensor!? key, int head_size,"
@@ -151,9 +145,6 @@ TORCH_LIBRARY_FRAGMENT(sgl_kernel, m) {
" float eps, float fp8_min, float fp8_max, bool scale_ue8m0, bool fuse_silu_and_mul, Tensor? masked_m) -> ()");
m.impl("sgl_per_token_group_quant_8bit_v2", torch::kCUDA, &sgl_per_token_group_quant_8bit_v2);
m.def("sgl_per_tensor_quant_fp8(Tensor input, Tensor! output_q, Tensor! output_s, bool is_static) -> ()");
m.impl("sgl_per_tensor_quant_fp8", torch::kCUDA, &sgl_per_tensor_quant_fp8);
m.def("sgl_per_token_quant_fp8(Tensor input, Tensor! output_q, Tensor! output_s) -> ()");
m.impl("sgl_per_token_quant_fp8", torch::kCUDA, &sgl_per_token_quant_fp8);
@@ -355,9 +346,6 @@ TORCH_LIBRARY_FRAGMENT(sgl_kernel, m) {
/*
* From csrc/memory
*/
m.def("store_kv_cache(Tensor k_cache, Tensor v_cache, Tensor out_loc, Tensor k, Tensor v) -> ()");
m.impl("store_kv_cache", &store_kv_cache);
m.def("weak_ref_tensor(Tensor tensor) -> Tensor");
m.impl("weak_ref_tensor", torch::kCUDA, &weak_ref_tensor);