Update on "Serialize NamedDataStoreOutput into PTD."

Update PTD serialization to account for blobs from the NamedDataStoreOutput.

Something we can do in the future is to consolidate tensors (that go through the emitter) and blobs (that come from the NamedDataStore).

Differential Revision: [D70939807](https://our.internmc.facebook.com/intern/diff/D70939807/)

[ghstack-poisoned]

Author: lucylq

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-295f2ed4d103017f7e19a7b8263ece606cd629db
+7ae0ce6360b6e4f944906502d20da24c04debee5

.ci/scripts/build_android_instrumentation.sh

@@ -12,30 +12,10 @@ if [[ -z "${PYTHON_EXECUTABLE:-}" ]]; then
 fi
 which "${PYTHON_EXECUTABLE}"
 
-build_android_test() {
-  mkdir -p extension/android/executorch_android/src/androidTest/resources
-  cp extension/module/test/resources/add.pte extension/android/executorch_android/src/androidTest/resources
-  pushd extension/android
-  ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:testDebugUnitTest
-  ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:assembleAndroidTest
-  popd
-}
+mkdir -p "${BUILD_AAR_DIR}"/executorch_android/src/androidTest/resources
+cp extension/module/test/resources/add.pte "${BUILD_AAR_DIR}"/executorch_android/src/androidTest/resources
 
-collect_artifacts_to_be_uploaded() {
-  ARTIFACTS_DIR_NAME="$1"
-  # Collect Java library test
-  JAVA_LIBRARY_TEST_DIR="${ARTIFACTS_DIR_NAME}/library_test_dir"
-  mkdir -p "${JAVA_LIBRARY_TEST_DIR}"
-  cp extension/android/executorch_android/build/outputs/apk/androidTest/debug/*.apk "${JAVA_LIBRARY_TEST_DIR}"
-}
-
-main() {
-  build_android_test
-  if [ -n "$ARTIFACTS_DIR_NAME" ]; then
-    collect_artifacts_to_be_uploaded ${ARTIFACTS_DIR_NAME}
-  fi
-}
-
-if [[ "${BASH_SOURCE[0]}" == "${0}" ]]; then
-  main "$@"
-fi
+pushd "${BUILD_AAR_DIR}"
+ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:testDebugUnitTest
+ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:assembleAndroidTest
+popd

.ci/scripts/gather_test_models.py

@@ -14,7 +14,7 @@
 from typing import Any
 
 from examples.models import MODEL_NAME_TO_MODEL
-from examples.xnnpack import MODEL_NAME_TO_OPTIONS
+from examples.xnnpack import MODEL_NAME_TO_OPTIONS, QuantType
 
 DEFAULT_RUNNERS = {
     "linux": "linux.2xlarge",
@@ -33,7 +33,7 @@
         "dl3": "linux.4xlarge.memory",
         "emformer_join": "linux.4xlarge.memory",
         "emformer_predict": "linux.4xlarge.memory",
-        "phi-4-mini": "linux.4xlarge.memory",
+        "phi_4_mini": "linux.4xlarge.memory",
     }
 }
 
@@ -154,7 +154,7 @@ def export_models_for_ci() -> dict[str, dict]:
         if backend == "xnnpack":
             if name not in MODEL_NAME_TO_OPTIONS:
                 continue
-            if MODEL_NAME_TO_OPTIONS[name].quantization:
+            if MODEL_NAME_TO_OPTIONS[name].quantization != QuantType.NONE:
                 backend += "-quantization"
 
             if MODEL_NAME_TO_OPTIONS[name].delegation:

.ci/scripts/test_llama_torchao_lowbit.sh

@@ -78,7 +78,6 @@ ${PYTHON_EXECUTABLE} -m examples.models.llama.export_llama \
     -qmode "torchao:8da${QLINEAR_BITWIDTH}w" \
     --group_size ${QLINEAR_GROUP_SIZE} \
     -E "torchao:${QEMBEDDING_BITWIDTH},${QEMBEDDING_GROUP_SIZE}" \
-    --disable_dynamic_shape \
     -d fp32
 
 # Test run

.ci/scripts/test_model.sh

@@ -100,11 +100,11 @@ test_model() {
       rm "./${MODEL_NAME}.pte"
       return  # Skip running with portable executor runnner since portable doesn't support Qwen's biased linears.
   fi
-  if [[ "${MODEL_NAME}" == "phi-4-mini" ]]; then
+  if [[ "${MODEL_NAME}" == "phi_4_mini" ]]; then
       # Install requirements for export_llama
       bash examples/models/llama/install_requirements.sh
       # Test export_llama script: python3 -m examples.models.llama.export_llama.
-      "${PYTHON_EXECUTABLE}" -m examples.models.llama.export_llama --model "${MODEL_NAME}" -c examples/models/llama/params/demo_rand_params.pth -p examples/models/phi-4-mini/config.json
+      "${PYTHON_EXECUTABLE}" -m examples.models.llama.export_llama --model "${MODEL_NAME}" -c examples/models/llama/params/demo_rand_params.pth -p examples/models/phi_4_mini/config.json
       run_portable_executor_runner
       rm "./${MODEL_NAME}.pte"
       return
@@ -224,19 +224,22 @@ test_model_with_coreml() {
 
   "${PYTHON_EXECUTABLE}" -m examples.apple.coreml.scripts.export --model_name="${MODEL_NAME}" --compute_precision "${DTYPE}"
   EXPORTED_MODEL=$(find "." -type f -name "${MODEL_NAME}*.pte" -print -quit)
-  # TODO:
+
   if [ -n "$EXPORTED_MODEL" ]; then
     EXPORTED_MODEL_WITH_DTYPE="${EXPORTED_MODEL%.pte}_${DTYPE}.pte"
     mv "$EXPORTED_MODEL" "$EXPORTED_MODEL_WITH_DTYPE"
     EXPORTED_MODEL="$EXPORTED_MODEL_WITH_DTYPE"
-    echo "Renamed file path: $EXPORTED_MODEL"
+    echo "OK exported model: $EXPORTED_MODEL"
   else
-    echo "No .pte file found"
+    echo "[error] failed to export model: no .pte file found"
     exit 1
   fi
 
   # Run the model
   if [ "${should_test}" = true ]; then
+    echo "Installing requirements needed to build coreml_executor_runner..."
+    backends/apple/coreml/scripts/install_requirements.sh
+
     echo "Testing exported model with coreml_executor_runner..."
     local out_dir=$(mktemp -d)
     COREML_EXECUTOR_RUNNER_OUT_DIR="${out_dir}" examples/apple/coreml/scripts/build_executor_runner.sh

.ci/scripts/unittest-linux.sh

@@ -21,8 +21,7 @@ if [[ "$BUILD_TOOL" == "cmake" ]]; then
     source .ci/scripts/setup-vulkan-linux-deps.sh
 
     PYTHON_EXECUTABLE=python \
-    EXECUTORCH_BUILD_PYBIND=ON \
-    CMAKE_ARGS="-DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
+    CMAKE_ARGS="-DEXECUTORCH_BUILD_PYBIND=ON -DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
     .ci/scripts/setup-linux.sh "$@"
 
     # Install llama3_2_vision dependencies.

.ci/scripts/unittest-macos-buck2.sh

@@ -21,8 +21,7 @@ trap 'rm -rfv ${TMP_DIR}' EXIT
 
 # Setup MacOS dependencies as there is no Docker support on MacOS atm
 PYTHON_EXECUTABLE=python \
-EXECUTORCH_BUILD_PYBIND=ON \
-CMAKE_ARGS="-DEXECUTORCH_BUILD_COREML=ON -DEXECUTORCH_BUILD_MPS=ON -DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
+CMAKE_ARGS="-DEXECUTORCH_BUILD_PYBIND=ON -DEXECUTORCH_BUILD_COREML=ON -DEXECUTORCH_BUILD_MPS=ON -DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
 ${CONDA_RUN} --no-capture-output \
 .ci/scripts/setup-macos.sh "$@"
 

.ci/scripts/unittest-macos.sh

@@ -60,12 +60,46 @@ install_pytorch_and_domains() {
   # Fetch the target commit
   pushd pytorch || return
   git checkout "${TORCH_VERSION}"
-  git submodule update --init --recursive
 
-  export USE_DISTRIBUTED=1
-  # Then build and install PyTorch
-  python setup.py bdist_wheel
-  pip install "$(echo dist/*.whl)"
+  local system_name=$(uname)
+  if [[ "${system_name}" == "Darwin" ]]; then
+    local platform=$(python -c 'import sysconfig; import platform; v=platform.mac_ver()[0].split(".")[0]; platform=sysconfig.get_platform().split("-"); platform[1]=f"{v}_0"; print("_".join(platform))')
+  fi
+  local python_version=$(python -c 'import platform; v=platform.python_version_tuple(); print(f"{v[0]}{v[1]}")')
+  local torch_release=$(cat version.txt)
+  local torch_short_hash=${TORCH_VERSION:0:7}
+  local torch_wheel_path="cached_artifacts/pytorch/executorch/pytorch_wheels/${system_name}/${python_version}"
+  local torch_wheel_name="torch-${torch_release}%2Bgit${torch_short_hash}-cp${python_version}-cp${python_version}-${platform:-}.whl"
+
+  local cached_torch_wheel="https://gha-artifacts.s3.us-east-1.amazonaws.com/${torch_wheel_path}/${torch_wheel_name}"
+  # Cache PyTorch wheel is only needed on MacOS, Linux CI already has this as part
+  # of the Docker image
+  local torch_wheel_not_found=0
+  if [[ "${system_name}" == "Darwin" ]]; then
+    pip install "${cached_torch_wheel}" || torch_wheel_not_found=1
+  else
+    torch_wheel_not_found=1
+  fi
+
+  # Found no such wheel, we will build it from source then
+  if [[ "${torch_wheel_not_found}" == "1" ]]; then
+    echo "No cached wheel found, continue with building PyTorch at ${TORCH_VERSION}"
+
+    git submodule update --init --recursive
+    USE_DISTRIBUTED=1 python setup.py bdist_wheel
+    pip install "$(echo dist/*.whl)"
+
+    # Only AWS runners have access to S3
+    if command -v aws && [[ -z "${GITHUB_RUNNER:-}" ]]; then
+      for wheel_path in dist/*.whl; do
+        local wheel_name=$(basename "${wheel_path}")
+        echo "Caching ${wheel_name}"
+        aws s3 cp "${wheel_path}" "s3://gha-artifacts/${torch_wheel_path}/${wheel_name}"
+      done
+    fi
+  else
+    echo "Use cached wheel at ${cached_torch_wheel}"
+  fi
 
   # Grab the pinned audio and vision commits from PyTorch
   TORCHAUDIO_VERSION=$(cat .github/ci_commit_pins/audio.txt)

.ci/scripts/utils.sh

@@ -8,13 +8,10 @@
 # should typically only contain shell variable assignments. Be sure to export
 # any variables so that subprocesses will see them.
 
-# Enable pybindings so that users can execute ExecuTorch programs from python.
-export EXECUTORCH_BUILD_PYBIND=1
-
 # Ensure that CMAKE_ARGS is defined before referencing it. Defaults to empty
 # if not defined.
 export CMAKE_ARGS="${CMAKE_ARGS:-}"
 
 # Link the XNNPACK backend into the pybindings runtime so that users can execute
 # ExecuTorch programs that delegate to it.
-CMAKE_ARGS="${CMAKE_ARGS} -DEXECUTORCH_BUILD_XNNPACK=ON"
+CMAKE_ARGS="${CMAKE_ARGS} -DEXECUTORCH_BUILD_PYBIND=ON -DEXECUTORCH_BUILD_XNNPACK=ON"

.ci/scripts/wheel/envvar_base.sh

@@ -15,11 +15,9 @@
                 model=Model.Mv3,
                 backend=Backend.XnnpackQuantizationDelegation,
             ),
-            # Enable this once CoreML is suppported out-of-the-box
-            # https://github.com/pytorch/executorch/issues/9019
-            # test_base.ModelTest(
-            #     model=Model.Mv3,
-            #     backend=Backend.CoreMlTest,
-            # )
+            test_base.ModelTest(
+                model=Model.Mv3,
+                backend=Backend.CoreMlTest,
+            ),
         ]
     )

.ci/scripts/wheel/test_macos.py

@@ -27,16 +27,21 @@ jobs:
         conda activate "${CONDA_ENV}"
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh --build-tool buck2
         export ARTIFACTS_DIR_NAME=artifacts-to-be-uploaded
-
-        mkdir -p ${ARTIFACTS_DIR_NAME}/fp32-xnnpack-custom
-        bash examples/models/llama/install_requirements.sh
-        bash ".ci/scripts/test_llama.sh" -model stories110M -build_tool cmake -dtype fp16 -mode portable -upload ${ARTIFACTS_DIR_NAME}/fp32-xnnpack-custom
+        mkdir -p ${ARTIFACTS_DIR_NAME}/
 
         # Build LLM Demo for Android
         export BUILD_AAR_DIR=aar-out
         mkdir -p $BUILD_AAR_DIR
-        bash scripts/build_android_library.sh ${ARTIFACTS_DIR_NAME}
-        bash .ci/scripts/build_android_instrumentation.sh ${ARTIFACTS_DIR_NAME}
+        bash scripts/build_android_library.sh
+        cp ${BUILD_AAR_DIR}/executorch.aar $ARTIFACTS_DIR_NAME
+
+        mkdir -p ${ARTIFACTS_DIR_NAME}/library_test_dir
+        bash .ci/scripts/build_android_instrumentation.sh
+        cp ${BUILD_AAR_DIR}/executorch_android/build/outputs/apk/androidTest/debug/executorch_android-debug-androidTest.apk "${ARTIFACTS_DIR_NAME}/library_test_dir"
+
+        mkdir -p ${ARTIFACTS_DIR_NAME}/fp32-xnnpack-custom
+        bash examples/models/llama/install_requirements.sh
+        bash ".ci/scripts/test_llama.sh" -model stories110M -build_tool cmake -dtype fp16 -mode portable -upload ${ARTIFACTS_DIR_NAME}/fp32-xnnpack-custom
 
         mkdir -p examples/demo-apps/android/LlamaDemo/app/libs
         cp aar-out/executorch.aar examples/demo-apps/android/LlamaDemo/app/libs
@@ -96,7 +101,7 @@ jobs:
           curl -O https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/llm_demo/app-debug.apk
           curl -O https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/llm_demo/app-debug-androidTest.apk
           curl -O https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/fp32-xnnpack-custom/model.zip
-          curl -o android-test-debug-androidTest.apk https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/library_test_dir/executorch-debug-androidTest.apk
+          curl -o android-test-debug-androidTest.apk https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/library_test_dir/executorch_android-debug-androidTest.apk
           unzip model.zip
           mv *.pte model.pte
 

.github/workflows/_android.yml

@@ -49,4 +49,6 @@ jobs:
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
       script: |
         set -eux
+        # This is needed to get the prebuilt PyTorch wheel from S3
+        ${CONDA_RUN} --no-capture-output pip install awscli==1.37.21
         .ci/scripts/unittest-macos.sh --build-tool "${{ inputs.build-tool }}" --build-mode "${{ inputs.build-mode }}" --editable "${{ inputs.editable }}"

.github/workflows/_unittest.yml

@@ -6,6 +6,9 @@ on:
     paths:
       - .ci/**/*
       - .github/workflows/build-wheels-linux.yml
+      - examples/**/*
+      - pyproject.toml
+      - setup.py
   push:
     branches:
       - nightly

.github/workflows/build-wheels-linux.yml

@@ -6,6 +6,9 @@ on:
     paths:
       - .ci/**/*
       - .github/workflows/build-wheels-macos.yml
+      - examples/**/*
+      - pyproject.toml
+      - setup.py
   push:
     branches:
       - nightly
@@ -57,6 +60,8 @@ jobs:
       pre-script: ${{ matrix.pre-script }}
       post-script: ${{ matrix.post-script }}
       package-name: ${{ matrix.package-name }}
-      runner-type: macos-m1-stable
+      # Meta's macOS runners do not have Xcode, so use GitHub's runners.
+      runner-type: macos-latest-xlarge
+      setup-miniconda: true
       smoke-test-script: ${{ matrix.smoke-test-script }}
       trigger-event: ${{ github.event_name }}

.github/workflows/build-wheels-macos.yml

@@ -106,7 +106,7 @@ jobs:
           - model: emformer_join
             backend: xnnpack-quantization-delegation
             runner: linux.4xlarge.memory
-          - model: phi-4-mini
+          - model: phi_4_mini
             backend: portable
             runner: linux.4xlarge.memory
           - model: llama3_2_vision_encoder
@@ -365,8 +365,7 @@ jobs:
         # build module for executorch.extension.pybindings.portable_lib
         BUILD_TOOL="cmake"
         PYTHON_EXECUTABLE=python \
-        EXECUTORCH_BUILD_XNNPACK=ON \
-        EXECUTORCH_BUILD_PYBIND=ON \
+        CMAKE_ARGS="-DEXECUTORCH_BUILD_PYBIND=ON -DEXECUTORCH_BUILD_XNNPACK=ON" \
         bash .ci/scripts/setup-linux.sh --build-tool "${BUILD_TOOL}"
 
         # see if we can import the module successfully
@@ -504,7 +503,7 @@ jobs:
 
         # Setup MacOS dependencies as there is no Docker support on MacOS atm
         PYTHON_EXECUTABLE=python \
-        EXECUTORCH_BUILD_PYBIND=ON \
+        CMAKE_ARGS="-DEXECUTORCH_BUILD_PYBIND=ON" \
         EXECUTORCH_BUILD_ARM_BAREMETAL=ON \
         .ci/scripts/setup-linux.sh --build-tool "${BUILD_TOOL}"
 

.github/workflows/pull.yml

@@ -65,22 +65,29 @@ jobs:
       matrix:
         model: [linear, add, add_mul, ic3, ic4, mv2, mv3, resnet18, resnet50, vit, w2l, mobilebert, emformer_join, emformer_transcribe]
         backend: [portable, xnnpack-quantization-delegation]
+        runner: [linux.arm64.2xlarge]
         include:
           - model: lstm
             backend: portable
+            runner: linux.arm64.2xlarge
           - model: mul
             backend: portable
+            runner: linux.arm64.2xlarge
           - model: softmax
             backend: portable
-          - model: phi-4-mini
+            runner: linux.arm64.2xlarge
+          - model: phi_4_mini
             backend: portable
+            runner: linux.arm64.m7g.4xlarge
           - model: qwen2_5
             backend: portable
+            runner: linux.arm64.2xlarge
           - model: llama3_2_vision_encoder
             backend: portable
+            runner: linux.arm64.2xlarge
       fail-fast: false
     with:
-      runner: linux.arm64.2xlarge
+      runner: ${{ matrix.runner }}
       docker-image: executorch-ubuntu-22.04-gcc11-aarch64
       submodules: 'true'
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
@@ -228,7 +235,7 @@ jobs:
     name: test-coreml-delegate
     uses: pytorch/test-infra/.github/workflows/macos_job.yml@main
     with:
-      runner: macos-13-xlarge
+      runner: macos-latest-xlarge
       python-version: '3.11'
       submodules: 'true'
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
@@ -261,7 +268,7 @@ jobs:
 
         # build module for executorch.extension.pybindings.portable_lib
         BUILD_TOOL=${{ matrix.build-tool }}
-        EXECUTORCH_BUILD_PYBIND=ON PYTHON_EXECUTABLE=python ${CONDA_RUN} bash .ci/scripts/setup-macos.sh --build-tool "${BUILD_TOOL}"
+        CMAKE_ARGS="-DEXECUTORCH_BUILD_PYBIND=ON" PYTHON_EXECUTABLE=python ${CONDA_RUN} bash .ci/scripts/setup-macos.sh --build-tool "${BUILD_TOOL}"
 
         # see if we can import the module successfully
         ${CONDA_RUN} python -c "from executorch.extension.pybindings import portable_lib; print('success!')"
@@ -536,9 +543,8 @@ jobs:
         git clone https://github.com/huggingface/optimum-executorch
         cd optimum-executorch
         # There is no release yet, for CI stability, always test from the same commit on main
-        git checkout 6a7e83f3eee2976fa809335bfb78a45b1ea1cb25
-        pip install .
-        pip install accelerate sentencepiece
+        git checkout 577a2b19670e4c643a5c6ecb09bf47b9a699e7c6
+        pip install .[tests]
         pip list
         echo "::endgroup::"