biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Sparse24] [Deprecation] Remove Sparse24 CT integration and kernels (#36799)

Browse Source 
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
This commit is contained in:
Kyle Sayers
2026-03-23 16:03:29 -04:00
committed by GitHub
parent fafe76b4af
commit 38364a7e32
17 changed files with 9 additions and 2674 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
csrc/torch_bindings.cpp
View File

@@ -523,26 +523,6 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
ops.impl("cutlass_scaled_mm_supports_block_fp8",
&cutlass_scaled_mm_supports_block_fp8);
// Check if cutlass sparse scaled_mm is supported for CUDA devices of the
// given capability
ops.def(
"cutlass_sparse_scaled_mm_supported(int cuda_device_capability) -> bool");
ops.impl("cutlass_sparse_scaled_mm_supported",
&cutlass_sparse_scaled_mm_supported);
// CUTLASS sparse GEMM, supporting symmetric per-tensor or per-row/column
// quantization, as well as bias
ops.def(
"cutlass_scaled_sparse_mm(Tensor! out, Tensor a,"
" Tensor bt_nzs,"
" Tensor bt_meta, Tensor a_scales,"
" Tensor b_scales, Tensor? bias) -> ()");
ops.impl("cutlass_scaled_sparse_mm", torch::kCUDA, &cutlass_scaled_sparse_mm);
// CUTLASS sparse matrix compressor
ops.def("cutlass_sparse_compress(Tensor a) -> Tensor[]");
ops.impl("cutlass_sparse_compress", &cutlass_sparse_compress);
// SM100 CUTLASS MLA decode
ops.def(
"sm100_cutlass_mla_decode(Tensor! out, Tensor! lse, Tensor q_nope,"