Merge remote-tracking branch 'meta_executorch/main' into dijopaul_nnlibdiv_patch

Merge back from meta main branch

Author: dijopaul

.ci/docker/build.sh

@@ -37,6 +37,10 @@ case "${IMAGE_NAME}" in
     ARM_SDK=yes
     CLANG_VERSION=12
     ;;
+  executorch-ubuntu-22.04-qnn-sdk)
+    QNN_SDK=yes
+    CLANG_VERSION=12
+    ;;
   executorch-ubuntu-22.04-clang12-android)
     LINTRUNNER=""
     CLANG_VERSION=12
@@ -72,6 +76,7 @@ docker build \
   --build-arg "LINTRUNNER=${LINTRUNNER:-}" \
   --build-arg "BUILD_DOCS=${BUILD_DOCS}" \
   --build-arg "ARM_SDK=${ARM_SDK:-}" \
+  --build-arg "QNN_SDK=${QNN_SDK:-}" \
   --build-arg "ANDROID_NDK_VERSION=${ANDROID_NDK_VERSION:-}" \
   -f "${OS}"/Dockerfile \
   "$@" \

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-d1b87e26e5c4343f5b56bb1e6f89b479b389bfac
+c8a648d4dffb9f0133ff4a2ea0e660b42105d3ad

.ci/docker/ci_commit_pins/torchao.txt

@@ -13,7 +13,7 @@ install_ubuntu() {
   apt-get install -y --no-install-recommends clang-"$CLANG_VERSION"
   apt-get install -y --no-install-recommends llvm-"$CLANG_VERSION"
   # Also require LLD linker from llvm and libomp to build PyTorch from source
-  apt-get install -y lld "libomp-${CLANG_VERSION}-dev"
+  apt-get install -y lld "libomp-${CLANG_VERSION}-dev" "libc++-${CLANG_VERSION}-dev"
 
   # Use update-alternatives to make this version the default
   update-alternatives --install /usr/bin/clang clang /usr/bin/clang-"$CLANG_VERSION" 50

.ci/docker/common/install_clang.sh

@@ -1,5 +1,5 @@
 mpmath==1.3.0
-numpy==1.22.0; python_version == '3.10'
+numpy==1.21.3; python_version == '3.10'
 numpy==1.23.2; python_version == '3.11'
 numpy; python_version >= '3.12'
 PyYAML==6.0.1

.ci/docker/requirements-ci.txt

@@ -82,5 +82,7 @@ COPY --chown=ci-user:ci-user ./arm /opt/arm
 # Set up ARM SDK if needed
 RUN if [ -n "${ARM_SDK}" ]; then git config --global user.email "[email protected]"; git config --global user.name "OSS CI"; bash /opt/arm/setup.sh --i-agree-to-the-contained-eula /opt/arm-sdk; chown -R ci-user:ci-user /opt/arm-sdk; fi
 
+ARG QNN_SDK
+
 USER ci-user
 CMD ["bash"]

.ci/docker/ubuntu/Dockerfile

@@ -88,8 +88,8 @@ def model_should_run_on_event(model: str, event: str) -> bool:
     if event == "pull_request":
         return model in ["mv3", "vit"]
     elif event == "push":
-        # 'emformer_predict' is running super slow. Only run it periodically
-        return model not in ["emformer_predict"]
+        # These are super slow. Only run it periodically
+        return model not in ["dl3", "edsr", "emformer_predict"]
     else:
         return True
 

.ci/scripts/gather_test_models.py

@@ -19,6 +19,8 @@ else
 fi
 
 # As Linux job is running inside a Docker container, all of its dependencies
-# have already been installed
-install_executorch
+# have already been installed, so we use PyTorch build from source here instead
+# of nightly. This allows CI to test against latest commits from PyTorch
+install_executorch "use-pt-pinned-commit"
 build_executorch_runner "${BUILD_TOOL}"
+do_not_use_nightly_on_ci

.ci/scripts/setup-linux.sh

@@ -31,8 +31,9 @@ install_qnn() {
 }
 
 setup_libc++() {
+  clang_version=$1
   sudo apt-get update
-  pkgs_to_check=('libc++-dev')
+  pkgs_to_check=("libc++-${clang_version}-dev")
   j=0
   while [ $j -lt ${#pkgs_to_check[*]} ]; do
     install_status=$(verify_pkg_installed ${pkgs_to_check[$j]})
@@ -47,5 +48,6 @@ setup_libc++() {
   done
 }
 
-setup_libc++
+# This needs to match with the clang version from the Docker image
+setup_libc++ 12
 install_qnn

.ci/scripts/setup-qnn-deps.sh

@@ -42,11 +42,12 @@ run_and_verify() {
 	-d fp32 \
 	--max_seq_length 32 \
 	--temperature 0 \
+    --show_tokens \
 	--prompt "Once upon a time," > result.txt
 
     # Verify result.txt
     RESULT=$(cat result.txt)
-    EXPECTED_RESULT="there was a little girl"
+    EXPECTED_RESULT="727, 471, 263, 2217, 7826, 4257, 365, 2354, 29889, 2296, 18012, 304, 1708, 5377, 297, 278, 6575, 845, 457, 29889, 3118, 2462, 29892, 1183, 4446, 263"
     if [[ "${RESULT}" == *"${EXPECTED_RESULT}"* ]]; then
         echo "Actual result: ${RESULT}"
         echo "Success"

.ci/scripts/test_llama_runner_eager.sh

@@ -20,8 +20,11 @@ install_executorch() {
   which pip
   # Install executorch, this assumes that Executorch is checked out in the
   # current directory.
-  # TODO(T199538337): clean up install scripts to use install_requirements.sh
-  ./install_requirements.sh --pybind xnnpack
+  if [[ "${1:-}" == "use-pt-pinned-commit" ]]; then
+    ./install_requirements.sh --pybind xnnpack --use-pt-pinned-commit
+  else
+    ./install_requirements.sh --pybind xnnpack
+  fi
   # Just print out the list of packages for debugging
   pip list
 }
@@ -110,10 +113,26 @@ cmake_install_executorch_lib() {
 }
 
 download_stories_model_artifacts() {
-    # Download stories110M.pt and tokenizer from Github
+  # Download stories110M.pt and tokenizer from Github
   curl -Ls "https://huggingface.co/karpathy/tinyllamas/resolve/main/stories110M.pt" --output stories110M.pt
   curl -Ls "https://raw.githubusercontent.com/karpathy/llama2.c/master/tokenizer.model" --output tokenizer.model
   # Create params.json file
   touch params.json
   echo '{"dim": 768, "multiple_of": 32, "n_heads": 12, "n_layers": 12, "norm_eps": 1e-05, "vocab_size": 32000}' > params.json
 }
+
+do_not_use_nightly_on_ci() {
+  # An assert to make sure that we are not using PyTorch nightly on CI to prevent
+  # regression as documented in https://github.com/pytorch/executorch/pull/6564
+  TORCH_VERSION=$(pip list | grep -w 'torch ' | awk -F ' ' {'print $2'} | tr -d '\n')
+
+  # The version of PyTorch building from source looks like 2.6.0a0+gitc8a648d that
+  # includes the commit while nightly (2.6.0.dev20241019+cpu) or release (2.6.0)
+  # won't have that. Note that we couldn't check for the exact commit from the pin
+  # ci_commit_pins/pytorch.txt here because the value will be different when running
+  # this on PyTorch CI
+  if [[ "${TORCH_VERSION}" != *"+git"* ]]; then
+    echo "Unexpected torch version. Expected binary built from source, got ${TORCH_VERSION}"
+    exit 1
+  fi
+}

.ci/scripts/utils.sh

@@ -0,0 +1,9 @@
+### Summary
+[PLEASE REMOVE] See [CONTRIBUTING.md's Pull Requests](https://github.com/pytorch/executorch/blob/main/CONTRIBUTING.md#pull-requests) for ExecuTorch PR guidelines.
+
+[PLEASE REMOVE] If this PR closes an issue, please add a `Fixes #<issue-id>` line.
+
+[PLEASE REMOVE] If this PR introduces a fix or feature that should be the upcoming release notes, please add a "Release notes: <area>" label. For a list of available release notes labels, check out [CONTRIBUTING.md's Pull Requests](https://github.com/pytorch/executorch/blob/main/CONTRIBUTING.md#pull-requests).
+
+### Test plan
+[PLEASE REMOVE] How did you test this PR? Please write down any manual commands you used and note down tests that you have written if applicable.

.github/pull_request_template.md

@@ -1,28 +1,9 @@
 name: Android
 
 on:
-  push:
-    branches:
-      - main
-      - release/*
-    tags:
-      - ciflow/android/*
-  pull_request:
-    paths:
-      - .ci/docker/**
-      - .github/workflows/android.yml
-      - build/*android*.sh
-      - install_requirements.sh
-      - examples/demo-apps/android/**
-      - extension/android/**
-      - extension/benchmark/android/**
-      - extension/module/**
+  workflow_call:
   workflow_dispatch:
 
-concurrency:
-  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}
-  cancel-in-progress: true
-
 jobs:
   build-llm-demo:
     name: build-llm-demo
@@ -85,6 +66,18 @@ jobs:
           # avoid permission issue
           sudo chown -R "${USER}" /opt/android
 
+      - name: Download Artifacts
+        shell: bash
+        run: |
+          set -eux
+          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.apk https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/library_test_dir/executorch-debug.apk
+          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
+          unzip model.zip
+          mv *.pte model.pte
+
       - name: Gradle cache
         uses: gradle/actions/setup-gradle@v3
 

.github/workflows/android.yml renamed to .github/workflows/_android.yml

@@ -37,6 +37,9 @@ jobs:
         CMAKE_ARGS="-DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
         .ci/scripts/setup-linux.sh cmake
 
+        # Install llama3_2_vision dependencies.
+        PYTHON_EXECUTABLE=python ./examples/models/llama3_2_vision/install_requirements.sh
+
         # Run pytest with coverage
         pytest -n auto --cov=./ --cov-report=xml
         # Run gtest
@@ -67,6 +70,10 @@ jobs:
         ${CONDA_RUN} --no-capture-output \
         .ci/scripts/setup-macos.sh cmake
 
+        # Install llama3_2_vision dependencies.
+        PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \
+        ./examples/models/llama3_2_vision/install_requirements.sh
+
         # Run pytest with coverage
         ${CONDA_RUN} pytest -n auto --cov=./ --cov-report=xml
         # Run gtest

.github/workflows/_unittest.yml

@@ -76,7 +76,7 @@ jobs:
           # on-demand and periodic benchmarking.
           CRON_DEFAULT_MODELS: "stories110M,mv3,mv2,ic4,ic3,resnet50,edsr,mobilebert,w2l"
           CRON_DEFAULT_DEVICES: "apple_iphone_15"
-          CRON_DEFAULT_DELEGATES: "nnpack,coreml,mps"
+          CRON_DEFAULT_DELEGATES: "xnnpack,coreml,mps"
         run: |
           set -ex
           MODELS="${{ inputs.models }}"

.github/workflows/apple-perf.yml

@@ -7,7 +7,6 @@ on:
       - release/*
   pull_request:
     paths:
-      - .ci/docker/**
       - .ci/scripts/setup-ios.sh
       - .github/workflows/apple.yml
       - install_requirements.sh

.github/workflows/apple.yml

@@ -38,6 +38,7 @@ jobs:
           - docker-image-name: executorch-ubuntu-22.04-clang12
           - docker-image-name: executorch-ubuntu-22.04-linter
           - docker-image-name: executorch-ubuntu-22.04-arm-sdk
+          - docker-image-name: executorch-ubuntu-22.04-qnn-sdk
           - docker-image-name: executorch-ubuntu-22.04-clang12-android
     env:
       DOCKER_IMAGE: 308535385114.dkr.ecr.us-east-1.amazonaws.com/executorch/${{ matrix.docker-image-name }}

.github/workflows/docker-builds.yml

@@ -5,16 +5,19 @@ on:
     branches:
       - 'gh/cccclai/[0-9]+/base'
       - 'gh/dbort/[0-9]+/base'
+      - 'gh/dvorjackz/[0-9]+/base'
       - 'gh/guangy10/[0-9]+/base'
       - 'gh/helunwencser/[0-9]+/base'
       - 'gh/jorgep31415/[0-9]+/base'
       - 'gh/kimishpatel/[0-9]+/base'
       - 'gh/kirklandsign/[0-9]+/base'
       - 'gh/larryliu0820/[0-9]+/base'
+      - 'gh/lucylq/[0-9]+/base'
       - 'gh/manuelcandales/[0-9]+/base'
       - 'gh/mcr229/[0-9]+/base'
       - 'gh/swolchok/[0-9]+/base'
       - 'gh/SS-JIA/[0-9]+/base'
+      - 'gh/trivedivivek/[0-9]+/base'
 
 jobs:
   ghstack_merge_to_main: