Fix wheel build

Author: larryliu0820

setup.py

@@ -814,8 +814,8 @@ def run(self):  # noqa C901
         if cmake_cache.is_enabled("EXECUTORCH_BUILD_PYBIND"):
             cmake_build_args += ["--target", "portable_lib"]
             cmake_build_args += ["--target", "selective_build"]
-        # TODO(larryliu0820): Temporarily disable building llm_runner for Windows
-        if cmake_cache.is_enabled("EXECUTORCH_BUILD_LLM_RUNNER") and not _is_windows():
+
+        if cmake_cache.is_enabled("EXECUTORCH_BUILD_EXTENSION_LLM_RUNNER"):
             cmake_build_args += ["--target", "_llm_runner"]
 
         if cmake_cache.is_enabled("EXECUTORCH_BUILD_EXTENSION_MODULE"):
@@ -890,7 +890,7 @@ def run(self):  # noqa C901
         BuiltExtension(
             src="extension/llm/runner/_llm_runner.*",  # @lint-ignore https://github.com/pytorch/executorch/blob/cb3eba0d7f630bc8cec0a9cc1df8ae2f17af3f7a/scripts/lint_xrefs.sh
             modpath="executorch.extension.llm.runner._llm_runner",
-            dependent_cmake_flags=["EXECUTORCH_BUILD_PYBIND"],
+            dependent_cmake_flags=["EXECUTORCH_BUILD_EXTENSION_LLM_RUNNER"],
         ),
         BuiltExtension(
             src="executorchcoreml.*",

tools/cmake/preset/pybind.cmake

@@ -13,8 +13,6 @@ set_overridable_option(EXECUTORCH_BUILD_KERNELS_QUANTIZED_AOT ON)
 set_overridable_option(EXECUTORCH_ENABLE_LOGGING ON)
 set_overridable_option(EXECUTORCH_LOG_LEVEL Info)
 set_overridable_option(EXECUTORCH_BUILD_XNNPACK ON)
-set_overridable_option(EXECUTORCH_BUILD_EXTENSION_LLM ON)
-set_overridable_option(EXECUTORCH_BUILD_EXTENSION_LLM_RUNNER ON)
 set_overridable_option(EXECUTORCH_BUILD_EXTENSION_TENSOR ON)
 set_overridable_option(EXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL ON)
 set_overridable_option(EXECUTORCH_BUILD_KERNELS_LLM ON)
@@ -24,12 +22,18 @@ set_overridable_option(EXECUTORCH_BUILD_EXTENSION_DATA_LOADER ON)
 set_overridable_option(EXECUTORCH_BUILD_KERNELS_OPTIMIZED ON)
 set_overridable_option(EXECUTORCH_BUILD_EXTENSION_MODULE ON)
 
+# TODO(larryliu0820): Temporarily disable building llm_runner for Windows
+# wheel due to the issue of tokenizer file path length limitation.
 if(CMAKE_SYSTEM_NAME STREQUAL "Darwin")
   set_overridable_option(EXECUTORCH_BUILD_COREML ON)
   set_overridable_option(EXECUTORCH_BUILD_EXTENSION_TRAINING ON)
+  set_overridable_option(EXECUTORCH_BUILD_EXTENSION_LLM_RUNNER ON)
+  set_overridable_option(EXECUTORCH_BUILD_EXTENSION_LLM ON)
 elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
   set_overridable_option(EXECUTORCH_BUILD_COREML ON)
   set_overridable_option(EXECUTORCH_BUILD_EXTENSION_TRAINING ON)
+  set_overridable_option(EXECUTORCH_BUILD_EXTENSION_LLM_RUNNER ON)
+  set_overridable_option(EXECUTORCH_BUILD_EXTENSION_LLM ON)
 elseif(CMAKE_SYSTEM_NAME STREQUAL "Windows" OR CMAKE_SYSTEM_NAME STREQUAL
                                                "WIN32"
 )