Merge branch 'main' into refactor-bitwise-logical-tests

Author: YufengShi-dudu

.ci/scripts/setup-openvino.sh

@@ -0,0 +1,28 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+set -ex
+
+# shellcheck source=/dev/null
+source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
+
+git clone https://github.com/openvinotoolkit/openvino.git
+cd openvino && git checkout releases/2025/1
+git submodule update --init --recursive
+sudo ./install_build_dependencies.sh
+mkdir build && cd build
+cmake .. -DCMAKE_BUILD_TYPE=Release -DENABLE_PYTHON=ON
+make -j$(nproc)
+
+cd ..
+cmake --install build --prefix dist
+
+source dist/setupvars.sh
+cd ../backends/openvino
+pip install -r requirements.txt
+cd scripts
+./openvino_build.sh --enable_python

.ci/scripts/test_llama.sh

@@ -269,7 +269,7 @@ $PYTHON_EXECUTABLE -m examples.models.llama.export_llama ${EXPORT_ARGS}
 
 # Create tokenizer.bin.
 echo "Creating tokenizer.bin"
-$PYTHON_EXECUTABLE -m extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
+$PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
 
 
 RUNTIME_ARGS="--model_path=${EXPORTED_MODEL_NAME} --tokenizer_path=tokenizer.bin --prompt=Once --temperature=0 --seq_len=10 --warmup=1"

.ci/scripts/test_llama_torchao_lowbit.sh

@@ -55,7 +55,7 @@ cmake --build cmake-out/examples/models/llama -j16 --config Release
 download_stories_model_artifacts
 
 echo "Creating tokenizer.bin"
-$PYTHON_EXECUTABLE -m extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
+$PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
 
 # Export model
 LLAMA_CHECKPOINT=stories110M.pt

.ci/scripts/test_openvino.sh

@@ -0,0 +1,16 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+set -ex
+
+# shellcheck source=/dev/null
+source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
+
+source openvino/dist/setupvars.sh
+cd backends/openvino/tests
+python test_runner.py --test_type ops
+python test_runner.py --test_type models

.ci/scripts/test_phi_3_mini.sh

@@ -56,7 +56,7 @@ cmake_build_phi_3_mini() {
 prepare_tokenizer() {
   echo "Downloading and converting tokenizer.model"
   wget -O tokenizer.model "https://huggingface.co/microsoft/Phi-3-mini-128k-instruct/resolve/main/tokenizer.model?download=true"
-  $PYTHON_EXECUTABLE -m executorch.extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
+  $PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
 }
 
 # Export phi-3-mini model to pte

.ci/scripts/test_qnn_static_llama.sh

@@ -30,7 +30,7 @@ pip install graphviz
 # Download stories llama110m artifacts
 download_stories_model_artifacts
 echo "Creating tokenizer.bin"
-$PYTHON_EXECUTABLE -m extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
+$PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
 
 set +e
 # Compile only as weight sharing is not applicable on x86

.github/workflows/pull.yml

@@ -736,3 +736,25 @@ jobs:
         conda activate "${CONDA_ENV}"
 
         # placeholder for mediatek to add more tests
+
+  test-openvino-linux:
+    name: test-openvino-linux
+    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
+    permissions:
+      id-token: write
+      contents: read
+    strategy:
+      fail-fast: false
+    with:
+      runner: linux.2xlarge
+      docker-image: executorch-ubuntu-22.04-gcc9
+      submodules: 'true'
+      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      timeout: 90
+      script: |
+        # The generic Linux job chooses to use base env, not the one setup by the image
+        CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
+        conda activate "${CONDA_ENV}"
+
+        PYTHON_EXECUTABLE=python bash .ci/scripts/setup-openvino.sh
+        PYTHON_EXECUTABLE=python bash .ci/scripts/test_openvino.sh

.gitmodules

@@ -1,9 +1,9 @@
 [submodule "backends/arm/third-party/ethos-u-core-driver"]
 	path = backends/arm/third-party/ethos-u-core-driver
-	url = https://github.com/pytorch-labs/ethos-u-core-driver-mirror
+	url = https://git.gitlab.arm.com/artificial-intelligence/ethos-u/ethos-u-core-driver.git
 [submodule "backends/arm/third-party/serialization_lib"]
 	path = backends/arm/third-party/serialization_lib
-	url = https://github.com/pytorch-labs/tosa_serialization_lib-mirror
+	url = https://git.gitlab.arm.com/tosa/tosa-serialization.git
 [submodule "backends/vulkan/third-party/Vulkan-Headers"]
 	path = backends/vulkan/third-party/Vulkan-Headers
 	url = https://github.com/KhronosGroup/Vulkan-Headers

.lintrunner.toml

@@ -299,12 +299,14 @@ include_patterns = [
     # TODO(https://github.com/pytorch/executorch/issues/7441): Gradually start enabling all folders.
     # 'backends/**/*.py',
     'backends/arm/**/*.py',
+    'backends/openvino/**/*.py',
     'build/**/*.py',
     'codegen/**/*.py',
     # 'devtools/**/*.py',
     'devtools/visualization/**/*.py',
     'docs/**/*.py',
     # 'examples/**/*.py',
+    'examples/openvino/**/*.py',
     # 'exir/**/*.py',
     # 'extension/**/*.py',
     'kernels/**/*.py',

CMakeLists.txt

@@ -1,6 +1,6 @@
 # Copyright (c) Meta Platforms, Inc. and affiliates.
-# Copyright 2024-2025 Arm Limited and/or its affiliates.
 # All rights reserved.
+# Copyright 2024-2025 Arm Limited and/or its affiliates.
 #
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
@@ -161,7 +161,7 @@ if(OPTIMIZE_SIZE)
   set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -Os")
 else()
   # -O2: Moderate opt.
-  set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -O2")
+  set(CMAKE_CXX_FLAGS_RELEASE "-O2 ${CMAKE_CXX_FLAGS_RELEASE}")
 endif()
 
 option(EXECUTORCH_BUILD_ANDROID_JNI "Build Android JNI" OFF)
@@ -204,6 +204,8 @@ option(EXECUTORCH_BUILD_MPS "Build the MPS backend" OFF)
 
 option(EXECUTORCH_BUILD_NEURON "Build the backends/mediatek directory" OFF)
 
+option(EXECUTORCH_BUILD_OPENVINO "Build the Openvino backend" OFF)
+
 option(EXECUTORCH_BUILD_PYBIND "Build the Python Bindings" OFF)
 
 option(EXECUTORCH_BUILD_QNN "Build the Qualcomm backend" OFF)
@@ -715,6 +717,10 @@ if(EXECUTORCH_BUILD_NEURON)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/backends/mediatek)
 endif()
 
+if(EXECUTORCH_BUILD_OPENVINO)
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/backends/openvino)
+endif()
+
 if(EXECUTORCH_BUILD_QNN)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/backends/qualcomm)
 endif()
@@ -751,7 +757,7 @@ if(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR)
 endif()
 
 if(EXECUTORCH_BUILD_EXTENSION_LLM)
-  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/llm/tokenizer)
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/llm/tokenizers)
 endif()
 
 if(EXECUTORCH_BUILD_EXTENSION_MODULE)
@@ -817,6 +823,10 @@ if(EXECUTORCH_BUILD_PYBIND)
     list(APPEND _dep_libs mpsdelegate)
   endif()
 
+  if(EXECUTORCH_BUILD_OPENVINO)
+    list(APPEND _dep_libs openvino_backend)
+  endif()
+
   if(EXECUTORCH_BUILD_XNNPACK)
     # need to explicitly specify XNNPACK and microkernels-prod
     # here otherwise uses XNNPACK and microkernel-prod symbols from libtorch_cpu