diff --git a/cmake/external_projects/tml_fa4.cmake b/cmake/external_projects/tml_fa4.cmake
index a965295e2b..c6ddff82c3 100644
--- a/cmake/external_projects/tml_fa4.cmake
+++ b/cmake/external_projects/tml_fa4.cmake
@@ -14,7 +14,7 @@ else()
   FetchContent_Declare(
     tml_fa4
     GIT_REPOSITORY https://github.com/vllm-project/tml-fa4.git
-    GIT_TAG b206834606ed5b5f21f8eed6b0683f528ea9cf7d
+    GIT_TAG 75765e76a9c2c012c1f6ecd64577eb646eb4d303
     GIT_PROGRESS TRUE
     CONFIGURE_COMMAND ""
     BUILD_COMMAND "")
diff --git a/requirements/cuda.txt b/requirements/cuda.txt
index 427f86ddbe..00a56ac3bd 100644
--- a/requirements/cuda.txt
+++ b/requirements/cuda.txt
@@ -26,8 +26,8 @@ fastsafetensors >= 0.3.3
 instanttensor >= 0.1.9
 
 # QuACK and Cutlass DSL for FA4 (cute-DSL implementation)
-nvidia-cutlass-dsl[cu13]==4.6.2
-quack-kernels==0.6.4  # Required for CUTLASS DSL 4.6 by MSA
+nvidia-cutlass-dsl[cu13]==4.7.1
+quack-kernels==0.6.5  # Required by MSA
 
 # Tokenspeed_MLA for faster mla with spec decode
 tokenspeed-mla==0.1.8; platform_system == "Linux"