vllm/v1/attention/backends/mla/flashmla_sparse.py · 7b01d97a22c977aaad2f38fa03cd11742d2893e8 · OpenDAS / vllm_cscc

[Attention][Perf][Kernel] Replace torch.cat with vectorized CUDA kernel MLA... · 580864d8

Roberto L. Castro authored Mar 09, 2026


[Attention][Perf][Kernel] Replace torch.cat with vectorized CUDA kernel MLA query concat - DeepSeek-V3.2 (#34917)
Signed-off-by: LopezCastroRoberto <rocastro@redhat.com>
Signed-off-by: Roberto L. Castro <38211239+LopezCastroRoberto@users.noreply.github.com>

580864d8

flashmla_sparse.py 33.4 KB

Replace flashmla_sparse.py