[AMD] Enable FP8 KV cache and FP8 attention kernel for NSA on MI300/MI355 with TileLang backend (#21511)

This commit is contained in:
Thomas Wang
2026-04-03 00:58:23 -07:00
committed by GitHub
parent ad0516d9c1
commit 7431db7392
6 changed files with 517 additions and 77 deletions
+1 -1
View File
@@ -88,7 +88,7 @@ ARG MOONCAKE_REPO="https://github.com/kvcache-ai/Mooncake.git"
ARG MOONCAKE_COMMIT="b6a841dc78c707ec655a563453277d969fb8f38d"
ARG TILELANG_REPO="https://github.com/tile-ai/tilelang.git"
ARG TILELANG_COMMIT="ebf4a7cb8881432165ae8760e99d209d905c704a"
ARG TILELANG_COMMIT="a55a82302bf7f3c5af635b5c9146f728185cc900"
ARG FHT_REPO="https://github.com/jeffdaily/fast-hadamard-transform.git"
ARG FHT_BRANCH="rocm"