[attn backend] Integrate tokenspeed_mla prefill/decode kernels (fp8 kv cache, blackwell) (#24925)

2026-05-13 17:36:17 -07:00
parent 22d3f3996c
commit 7618ad7075
11 changed files with 462 additions and 92 deletions
@@ -65,6 +65,7 @@ dependencies = [
  "tiktoken",
  "tilelang==0.1.8",
  "timm==1.0.16",
+  "tokenspeed_mla==0.1.1",
  "torch_memory_saver>=0.0.9.post1",
  "torch==2.11.0",
  "torchao==0.17.0",