up

Author: metascroy

.github/workflows/trunk.yml

@@ -387,7 +387,7 @@ jobs:
         eval "$(conda shell.bash hook)"
 
         # Install requirements
-        ${CONDA_RUN} ET_BUILD_TORCHAO_KERNELS=1 python install_executorch.py
+        ${CONDA_RUN} EXECUTORCH_BUILD_TORCHAO=1 python install_executorch.py
         ${CONDA_RUN} sh examples/models/llama/install_requirements.sh
 
         # Run test

examples/models/llama/README.md

@@ -339,9 +339,9 @@ Please refer to [this tutorial](https://pytorch.org/executorch/main/llm/llama-de
 
 ## Running with low-bit kernels
 
-We now give instructions for quantizating and running your model with low-bit kernels.  These are still experimental, and require you do development on an Arm-based Mac, and install executorch from source with the environment variable ET_BUILD_TORCHAO_KERNELS=1 defined:
+We now give instructions for quantizating and running your model with low-bit kernels.  These are still experimental, and require you do development on an Arm-based Mac, and install executorch from source with the environment variable EXECUTORCH_BUILD_TORCHAO=1 defined:
 ```
-ET_BUILD_TORCHAO_KERNELS=1 python install_executorch.py
+EXECUTORCH_BUILD_TORCHAO=1 python install_executorch.py
 ```
 
 Also note that low-bit quantization often requires QAT (quantization-aware training) to give good quality results.

install_requirements.py

@@ -118,13 +118,14 @@ def install_requirements(use_pytorch_nightly):
     # Install packages directly from local copy instead of pypi.
     # This is usually not recommended.
     new_env = os.environ.copy()
-    if ("ET_BUILD_TORCHAO_KERNELS" not in new_env) or (
-        new_env["ET_BUILD_TORCHAO_KERNELS"] == "0"
+    if ("EXECUTORCH_BUILD_TORCHAO" not in new_env) or (
+        new_env["EXECUTORCH_BUILD_TORCHAO"] == "0"
     ):
         new_env["USE_CPP"] = "0"
     else:
-        assert new_env["ET_BUILD_TORCHAO_KERNELS"] == "1"
+        assert new_env["EXECUTORCH_BUILD_TORCHAO"] == "1"
         new_env["USE_CPP"] = "1"
+        new_env["CMAKE_POLICY_VERSION_MINIMUM"] = "3.5"
     subprocess.run(
         [
             sys.executable,

tools/cmake/preset/llm.cmake

@@ -17,6 +17,9 @@ set_overridable_option(EXECUTORCH_BUILD_XNNPACK ON)
 if(CMAKE_SYSTEM_NAME STREQUAL "Darwin")
   set_overridable_option(EXECUTORCH_BUILD_COREML ON)
   set_overridable_option(EXECUTORCH_BUILD_MPS ON)
+  if(CMAKE_SYSTEM_PROCESSOR STREQUAL "arm64")
+    set_overridable_option(EXECUTORCH_BUILD_TORCHAO ON)
+  endif()
 elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
   # Linux-specific code here
 elseif(CMAKE_SYSTEM_NAME STREQUAL "Windows" OR CMAKE_SYSTEM_NAME STREQUAL "WIN32")