[MUSA][9/N] Add FA3 attention backend support through MATE (MUSA AI Tensor Engine) (#17985)

Co-authored-by: R0CKSTAR <xiaodong.ye@mthreads.com>
2026-04-02 15:04:31 -07:00
parent 566b4a4f1c
commit 939cf398a9
9 changed files with 536 additions and 47 deletions
@@ -116,6 +116,9 @@ srt_musa = [
    "torch_musa",
    "torchada>=0.1.45",
    "mthreads-ml-py",
+    "mate",
+    "mate-deep_gemm",
+    "mate-flash-attention",
    "numpy<2.0",
 ]