fa4 cleanup (#19727)

sgl-kernel/CMakeLists.txt

@@ -570,44 +570,3 @@ install(DIRECTORY "${repo-triton_SOURCE_DIR}/python/triton_kernels/triton_kernel
         DESTINATION "triton_kernels"
         PATTERN ".git*" EXCLUDE
         PATTERN "__pycache__" EXCLUDE)
-
-# ============================ Extra Install: FA4 ============================= #
-# TODO: find a better install condition.
-if ("${CUDA_VERSION}" VERSION_GREATER_EQUAL "12.8" OR SGL_KERNEL_ENABLE_SM100A)
-
-    set(FLASH_ATTN_CUTE_SRC "${repo-flash-attention_SOURCE_DIR}/flash_attn/cute")
-    set(FLASH_ATTN_CUTE_DST "${CMAKE_CURRENT_BINARY_DIR}/flash_attn_origin/cute")
-
-    file(MAKE_DIRECTORY "${FLASH_ATTN_CUTE_DST}")
-
-    file(COPY "${FLASH_ATTN_CUTE_SRC}/"
-         DESTINATION "${FLASH_ATTN_CUTE_DST}"
-         PATTERN ".git*" EXCLUDE
-         PATTERN "__pycache__" EXCLUDE)
-
-    file(GLOB_RECURSE FLASH_ATTN_CUTE_DST_PY
-         "${FLASH_ATTN_CUTE_DST}/*.py")
-
-    foreach(FILE_PATH IN LISTS FLASH_ATTN_CUTE_DST_PY)
-        file(READ "${FILE_PATH}" FILE_CONTENT)
-
-        set(MODIFIED_CONTENT "${FILE_CONTENT}")
-
-        # The main goal is to avoid using "flash_attn" so that other libraries (such as transformers) do not mistakenly assume that "flash_attn" is already installed.
-
-        string(REPLACE "flash_attn.cute"
-                       "flash_attn_origin.cute"
-                       MODIFIED_CONTENT "${MODIFIED_CONTENT}")
-
-        if (NOT FILE_CONTENT STREQUAL MODIFIED_CONTENT)
-            file(WRITE "${FILE_PATH}" "${MODIFIED_CONTENT}")
-            message(STATUS "  - [FA4 Patch] Patched: ${FILE_PATH}")
-        endif()
-    endforeach()
-
-    install(DIRECTORY "${FLASH_ATTN_CUTE_DST}/"
-            DESTINATION "flash_attn_origin/cute"
-            PATTERN ".git*" EXCLUDE
-            PATTERN "__pycache__" EXCLUDE)
-
-endif()