Fix includes: use cutlass/float_subbyte.h (has float_e2m1_t and float_ue4m3_t), point to latest CUTLASS

src/nvfp4_megamoe_kernel/cutlass_nvfp4_gemm/cutlass_nvfp4_gemm.cu

@@ -43,7 +43,7 @@
 #include "cutlass/kernel_hardware_info.hpp"
 #include "cutlass/detail/sm100_blockscaled_layout.hpp"
 
-#include "cute/numeric/float8.hpp"
+#include "cutlass/float_subbyte.h"
 #include "cute/layout.hpp"
 
 using namespace cute;

src/nvfp4_megamoe_kernel/cutlass_nvfp4_gemm/pytorch_binding.cpp

@@ -27,7 +27,7 @@
 #include "cutlass/kernel_hardware_info.hpp"
 #include "cutlass/detail/sm100_blockscaled_layout.hpp"
 
-#include "cute/numeric/float8.hpp"
+#include "cutlass/float_subbyte.h"
 #include "cute/layout.hpp"
 
 using namespace cute;

src/nvfp4_megamoe_kernel/cutlass_nvfp4_gemm/setup.py

@@ -1,18 +1,17 @@
-"""
-Setup script for CUTLASS NVFP4 block-scaled GEMM PyTorch extension.
-"""
+"""Setup script for CUTLASS NVFP4 block-scaled GEMM PyTorch extension."""
 
 import os
 from setuptools import setup
 from torch.utils.cpp_extension import BuildExtension, CUDAExtension
 
-# CUTLASS include directory
+# CUTLASS include directory — prefer the latest from GitHub
 CUTLASS_INCLUDE_DIR = os.environ.get(
     "CUTLASS_INCLUDE_DIR",
-    "/usr/local/lib/python3.12/dist-packages/tilelang/3rdparty/cutlass/include"
+    "/root/cutlass/include"
 )
 if not os.path.exists(os.path.join(CUTLASS_INCLUDE_DIR, "cutlass", "cutlass.h")):
     for alt in [
+        "/root/cutlass/include",
         "/usr/local/lib/python3.12/dist-packages/tilelang/3rdparty/cutlass/include",
         "/usr/local/include/cutlass",
         "/opt/cutlass/include",