[NVIDIA] Support Cutlass MLA for Blackwell GPUs (#16032)

Signed-off-by: kaixih <kaixih@nvidia.com>
This commit is contained in:
Kaixi Hou
2025-04-27 06:29:21 -07:00
committed by GitHub
parent 756848e79e
commit ed7a29d9f8
8 changed files with 403 additions and 5 deletions
+6
View File
@@ -128,6 +128,12 @@ void advance_step_flashinfer(
torch::Tensor& paged_kv_indices, torch::Tensor& paged_kv_indptr,
torch::Tensor& paged_kv_last_page_len, torch::Tensor& block_table_bounds);
void cutlass_mla_decode(torch::Tensor const& out, torch::Tensor const& q_nope,
torch::Tensor const& q_pe,
torch::Tensor const& kv_c_and_k_pe_cache,
torch::Tensor const& seq_lens,
torch::Tensor const& page_table, double scale);
torch::Tensor get_cuda_view_from_cpu_tensor(torch::Tensor& cpu_tensor);
#ifndef USE_ROCM