Merge branch 'main' into jz/phi-4-mini-readme

Author: jackzhxng

.ci/scripts/unittest-macos.sh

@@ -19,14 +19,14 @@ export TMP_DIR=$(mktemp -d)
 export PATH="${TMP_DIR}:$PATH"
 trap 'rm -rfv ${TMP_DIR}' EXIT
 
-if [[ "$BUILD_TOOL" == "cmake" ]]; then
-    # 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" \
-    ${CONDA_RUN} --no-capture-output \
-    .ci/scripts/setup-macos.sh "$@"
+# 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" \
+${CONDA_RUN} --no-capture-output \
+.ci/scripts/setup-macos.sh "$@"
 
+if [[ "$BUILD_TOOL" == "cmake" ]]; then
     # Install llama3_2_vision dependencies.
     PYTHON_EXECUTABLE=python \
     ${CONDA_RUN} --no-capture-output \

.github/workflows/_android.yml

@@ -35,7 +35,7 @@ jobs:
         # Build LLM Demo for Android
         export BUILD_AAR_DIR=aar-out
         mkdir -p $BUILD_AAR_DIR
-        bash build/build_android_library.sh ${ARTIFACTS_DIR_NAME}
+        bash scripts/build_android_library.sh ${ARTIFACTS_DIR_NAME}
         bash .ci/scripts/build_android_instrumentation.sh ${ARTIFACTS_DIR_NAME}
 
         mkdir -p examples/demo-apps/android/LlamaDemo/app/libs
@@ -120,7 +120,7 @@ jobs:
         with:
           api-level: ${{ env.API_LEVEL }}
           arch: x86_64
-          script: ./build/run_android_emulator.sh
+          script: ./scripts/run_android_emulator.sh
           # NB: This is to boot the emulator faster following the instructions on
           # https://github.com/ReactiveCircus/android-emulator-runner. The max number
           # of cores we can set is 6, any higher number will be reduced to 6.

.github/workflows/android-perf.yml

@@ -363,7 +363,7 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/build-qnn-sdk.sh
 
         mkdir -p aar-out
-        PYTHON_EXECUTABLE=python ANDROID_ABIS="arm64-v8a" BUILD_AAR_DIR=aar-out EXECUTORCH_BUILD_QNN=ON QNN_SDK_ROOT=/tmp/qnn/2.28.0.241029 bash build/build_android_library.sh
+        PYTHON_EXECUTABLE=python ANDROID_ABIS="arm64-v8a" BUILD_AAR_DIR=aar-out EXECUTORCH_BUILD_QNN=ON QNN_SDK_ROOT=/tmp/qnn/2.28.0.241029 bash scripts/build_android_library.sh
         mkdir -p extension/benchmark/android/benchmark/app/libs
         cp aar-out/executorch.aar extension/benchmark/android/benchmark/app/libs
         pushd extension/benchmark/android/benchmark

.github/workflows/android-release-artifacts.yml

@@ -55,7 +55,7 @@ jobs:
         # Build AAR Package
         mkdir aar-out
         export BUILD_AAR_DIR=aar-out
-        bash build/build_android_library.sh
+        bash scripts/build_android_library.sh
         mkdir -p "${ARTIFACTS_DIR_NAME}"
         cp aar-out/executorch.aar "${ARTIFACTS_DIR_NAME}/executorch.aar"
 

.github/workflows/apple-perf.yml

@@ -387,7 +387,7 @@ jobs:
 
         echo "::group::Build ExecuTorch iOS frameworks"
         PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \
-          build/build_apple_frameworks.sh --Release --Debug --coreml --custom --mps --optimized --portable --quantized --xnnpack
+          scripts/build_apple_frameworks.sh --Release --Debug --coreml --custom --mps --optimized --portable --quantized --xnnpack
         echo "::endgroup::"
 
         # NB: Although exported models can be copied to this directory and bundled together with the
@@ -396,7 +396,7 @@ jobs:
         # create the directory here to pass the build
         mkdir -p extension/benchmark/apple/Benchmark/Models
         ${CONDA_RUN} --no-capture-output \
-          build/build_apple_llm_demo.sh ${ARTIFACTS_DIR_NAME}
+          scripts/build_apple_llm_demo.sh ${ARTIFACTS_DIR_NAME}
 
   upload-benchmark-app:
     needs: build-benchmark-app

.github/workflows/apple.yml

@@ -11,9 +11,9 @@ on:
       - .github/workflows/apple.yml
       - install_executorch.sh
       - backends/apple/**
-      - build/build_apple_frameworks.sh
-      - build/build_apple_llm_demo.sh
-      - build/create_frameworks.sh
+      - scripts/build_apple_frameworks.sh
+      - scripts/build_apple_llm_demo.sh
+      - scripts/create_frameworks.sh
       - .ci/scripts/test_ios_ci.sh
       - examples/demo-apps/apple_ios/**
       - extension/apple/**
@@ -172,7 +172,7 @@ jobs:
 
         # Build iOS Frameworks
         PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \
-        build/build_apple_frameworks.sh --Release --Debug --coreml --custom --mps --optimized --portable --quantized --xnnpack
+        scripts/build_apple_frameworks.sh --Release --Debug --coreml --custom --mps --optimized --portable --quantized --xnnpack
 
         # Bundle iOS Frameworks
         for FRAMEWORK in "${FRAMEWORKS[@]}"; do (
@@ -313,11 +313,11 @@ jobs:
 
         echo "::group::Build ExecuTorch iOS frameworks"
         PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \
-          build/build_apple_frameworks.sh --Release --Debug --coreml --custom --mps --optimized --portable --quantized --xnnpack
+          scripts/build_apple_frameworks.sh --Release --Debug --coreml --custom --mps --optimized --portable --quantized --xnnpack
         echo "::endgroup::"
 
         echo "::group::Build ExecuTorch benchmark app"
         mkdir -p extension/benchmark/apple/Benchmark/Models
         ${CONDA_RUN} --no-capture-output \
-          build/build_apple_llm_demo.sh "${ARTIFACTS_DIR_NAME}"
+          scripts/build_apple_llm_demo.sh "${ARTIFACTS_DIR_NAME}"
         echo "::endgroup::"

.github/workflows/pull.yml

@@ -290,6 +290,36 @@ jobs:
         # run e2e (export, tokenizer and runner)
         PYTHON_EXECUTABLE=python bash .ci/scripts/test_llava.sh
 
+  test-moshi-linux:
+    name: test-moshi-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-clang12
+      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-linux.sh --build-tool "cmake"
+
+        # install Mimi requirements
+        bash examples/models/moshi/mimi/install_requirements.sh
+
+        # reinstall executorch
+        bash ./install_executorch.sh
+
+        # run python unittest
+        python -m unittest examples.models.moshi.mimi.test_mimi
+
   test-quantized-aot-lib-linux:
     name: test-quantized-aot-lib-linux
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main

.gitignore

@@ -11,6 +11,7 @@ __pycache__/
 arm_test/
 buck-out/
 buck2-bin/
+build/
 cmake-android-out/
 cmake-ios-out/
 cmake-out*

README.md

@@ -67,7 +67,6 @@ We welcome contributions. To get started review the [guidelines](CONTRIBUTING.md
 ```
 executorch
 ├── backends                        #  Backend delegate implementations.
-├── build                           #  Utilities for managing the build system.
 ├── codegen                         #  Tooling to autogenerate bindings between kernels and the runtime.
 ├── configurations
 ├── docs                            #  Static docs tooling.
@@ -81,8 +80,7 @@ executorch
 |   ├── operator                    #  Operator node manipulation utilities.
 |   ├── passes                      #  Built-in compiler passes.
 |   ├── program                     #  Export artifacts.
-|   ├── serde                       #  Graph module
-serialization/deserialization.
+|   ├── serde                       #  Graph module serialization/deserialization.
 |   ├── verification                #  IR verification.
 ├── extension                       #  Extensions built on top of the runtime.
 |   ├── android                     #  ExecuTorch wrappers for Android apps.
@@ -97,8 +95,7 @@ serialization/deserialization.
 |   ├── parallel                    #  C++ threadpool integration.
 |   ├── pybindings                  #  Python API for executorch runtime.
 |   ├── pytree                      #  C++ and Python flattening and unflattening lib for pytrees.
-|   ├── runner_util                 #  Helpers for writing C++ PTE-execution
-tools.
+|   ├── runner_util                 #  Helpers for writing C++ PTE-execution tools.
 |   ├── testing_util                #  Helpers for writing C++ tests.
 |   ├── training                    #  Experimental libraries for on-device training
 ├── kernels                         #  1st party kernel implementations.
@@ -114,9 +111,9 @@ tools.
 |   ├── executor                    #  Model loading, initialization, and execution.
 |   ├── kernel                      #  Kernel registration and management.
 |   ├── platform                    #  Layer between architecture specific code and portable C++.
-├── schema                          #  ExecuTorch PTE file format flatbuffer
-schemas.
-├── scripts                         #  Utility scripts for size management, dependency management, etc.
+├── schema                          #  ExecuTorch PTE file format flatbuffer schemas.
+├── scripts                         #  Utility scripts for building libs, size management, dependency management, etc.
+├── tools                           #  Development tool management.
 ├── devtools                        #  Model profiling, debugging, and introspection.
 ├── shim                            #  Compatibility layer between OSS and Internal builds
 ├── test                            #  Broad scoped end-to-end tests.

backends/apple/coreml/setup.md

@@ -50,7 +50,7 @@ xcode-select --install
 
 ```bash
 cd executorch
-./build/build_apple_frameworks.sh --coreml
+./scripts/build_apple_frameworks.sh --coreml
 ```
 5. Open the project in Xcode, and drag `executorch.xcframework` and `coreml_backend.xcframework` frameworks generated from Step 2 to Frameworks.
 

backends/apple/mps/setup.md

@@ -130,7 +130,7 @@ python3 -m sdk.inspector.inspector_cli --etdump_path etdump.etdp --etrecord_path
 ***Step 1***. Create the ExecuTorch core and MPS delegate frameworks to link on iOS
 ```bash
 cd executorch
-./build/build_apple_frameworks.sh --mps
+./scripts/build_apple_frameworks.sh --mps
 ```
 
 `mps_delegate.xcframework` will be in `cmake-out` folder, along with `executorch.xcframework` and `portable_delegate.xcframework`:

backends/arm/_passes/insert_table_ops.py

@@ -41,6 +41,7 @@ class InsertTableOpsPass(ExportPass):
     """
 
     table_ops: Dict[EdgeOpOverload, Callable[[torch.Tensor], torch.Tensor]] = {
+        exir_ops.edge.aten.ceil.default: torch.ceil,
         exir_ops.edge.aten.exp.default: torch.exp,
         exir_ops.edge.aten.floor.default: torch.floor,
         exir_ops.edge.aten.log.default: torch.log,

backends/arm/operator_support/tosa_supported_operators.py

@@ -147,6 +147,7 @@ def is_node_supported(
             exir_ops.edge.aten.bitwise_xor.Tensor,
             exir_ops.edge.aten.expand_copy.default,
             exir_ops.edge.aten.cat.default,
+            exir_ops.edge.aten.ceil.default,
             exir_ops.edge.aten.clamp.default,
             exir_ops.edge.aten.bmm.default,
             exir_ops.edge.aten.permute_copy.default,
@@ -208,26 +209,41 @@ def is_node_supported(
 class EthosU55NotSupported(OperatorSupportBase):
     """
     Certain operators are not supported on U55. These are listed in `unsupported_ops`.
+    The comment mentions the unsupported TOSA operator that the aten operator maps to where it is not obvious.
+    For unimplemented operators, this is the anticipated mapping, and it might be incorrect.
     """
 
     unsupported_ops = [
-        exir_ops.edge.aten.any.default,
-        exir_ops.edge.aten.any.dim,
-        exir_ops.edge.aten.any.dims,
+        exir_ops.edge.aten.any.default,  # REDUCE_ANY
+        exir_ops.edge.aten.any.dim,  # REDUCE_ANY
+        exir_ops.edge.aten.any.dims,  # REDUCE_ANY
         exir_ops.edge.aten.bitwise_and.Tensor,
         exir_ops.edge.aten.bitwise_or.Tensor,
         exir_ops.edge.aten.bitwise_xor.Tensor,
+        exir_ops.edge.aten.bitwise_not,
         exir_ops.edge.aten.logical_and.default,
         exir_ops.edge.aten.logical_or.default,
         exir_ops.edge.aten.logical_xor.default,
         exir_ops.edge.aten.logical_not.default,
-        exir_ops.edge.aten.amax.default,
-        exir_ops.edge.aten.amin.default,
+        exir_ops.edge.aten.amax.default,  # REDUCE_MAX
+        exir_ops.edge.aten.amin.default,  # REDUCE_MIN
         exir_ops.edge.aten.eq.Tensor,
         exir_ops.edge.aten.ge.Tensor,
         exir_ops.edge.aten.gt.Tensor,
         exir_ops.edge.aten.le.Tensor,
         exir_ops.edge.aten.lt.Tensor,
+        exir_ops.edge.aten.flip.default,  # REVERSE
+        exir_ops.edge.aten.grid_sampler_2d,  # GATHER
+        exir_ops.edge.aten.scatter.src,
+        exir_ops.edge.aten.scatter.value,
+        exir_ops.edge.aten.select_scatter.default,
+        exir_ops.edge.aten.scatter_reduce.two,
+        exir_ops.edge.aten.scatter_add.default,
+        exir_ops.edge.aten.upsample_nearest2d.vec,  # RESIZE
+        exir_ops.edge.aten.upsample_bilinear2d.vec,  # RESIZE
+        exir_ops.edge.aten.reflection_pad1d.default,  # REVERSE
+        exir_ops.edge.aten.reflection_pad2d.default,  # REVERSE
+        exir_ops.edge.aten.reflection_pad3d.default,  # REVERSE
     ]
 
     def __init__(self, reporter: WhyNoPartitionReporter):

backends/arm/operators/ops_unary.py

@@ -53,5 +53,6 @@ def define_node(
     register_node_visitor(UnaryOperator)
 
 
+unary_operator_factory("aten.ceil.default", TosaOp.Op().CEIL)
 unary_operator_factory("aten.floor.default", TosaOp.Op().FLOOR)
 unary_operator_factory("aten.logical_not.default", TosaOp.Op().LOGICAL_NOT)

backends/arm/quantizer/quantization_annotator.py

@@ -126,6 +126,7 @@ def _match_pattern(
 
 _one_to_one = [
     torch.ops.aten.abs.default,
+    torch.ops.aten.ceil.default,
     torch.ops.aten.exp.default,
     torch.ops.aten.floor.default,
     torch.ops.aten.log.default,

backends/arm/test/common.py

@@ -231,6 +231,7 @@ def parametrize(
     arg_name: str,
     test_data: dict[str, Any],
     xfails: dict[str, xfail_type] | None = None,
+    strict: bool = True,
 ):
     """
     Custom version of pytest.mark.parametrize with some syntatic sugar and added xfail functionality
@@ -261,7 +262,9 @@ def decorator_func(func):
                 pytest_param = pytest.param(
                     test_parameters,
                     id=id,
-                    marks=pytest.mark.xfail(reason=reason, raises=raises, strict=True),
+                    marks=pytest.mark.xfail(
+                        reason=reason, raises=raises, strict=strict
+                    ),
                 )
             else:
                 pytest_param = pytest.param(test_parameters, id=id)