Openvino llama support

.ci/docker/build.sh

@@ -54,13 +54,13 @@ case "${IMAGE_NAME}" in
   executorch-ubuntu-22.04-mediatek-sdk)
     MEDIATEK_SDK=yes
     CLANG_VERSION=12
-    ANDROID_NDK_VERSION=r27b
+    ANDROID_NDK_VERSION=r28c
     ;;
   executorch-ubuntu-22.04-clang12-android)
     LINTRUNNER=""
     CLANG_VERSION=12
     # From https://developer.android.com/ndk/downloads
-    ANDROID_NDK_VERSION=r27b
+    ANDROID_NDK_VERSION=r28c
     ;;
   *)
     echo "Invalid image name ${IMAGE_NAME}"

.ci/docker/ci_commit_pins/optimum-executorch.txt

@@ -1 +1 @@
-40b02a2dc61bbf901a2df91719f47c98d65368ec
+44d8d54e38c0258357d4e92e1fefe21e845947a3

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-4d4abec80f03cd8fdefe1d9cb3a60d3690cd777e
+cf9d09490c7f6685ec68d5db3acf2e0d73c54d00

.ci/docker/requirements-ci.txt

@@ -16,18 +16,21 @@ hypothesis==6.84.2
 parameterized==0.9.0
 
 # Doc build requirements, same as https://github.com/pytorch/pytorch/blob/main/.ci/docker/requirements-docs.txt
-sphinx==5.3.0
+sphinx==7.2.6
+sphinxcontrib.katex==0.9.10
+breathe==4.36.0  # only if generating C++
+exhale==0.3.7  # only if generating C++ docs
+docutils==0.18.1,<0.21
+sphinx-design==0.6.1
+sphinxcontrib-mermaid==1.0.0
+myst-parser==3.0.1  # if want to contribute in markdown
+sphinx-gallery==0.14.0  # only if hosting interactive tutorials
+sphinx-sitemap==2.7.1
 sphinx-reredirects==0.1.4
-sphinx-gallery==0.14.0
-breathe==4.34.0
-exhale==0.2.3
-docutils==0.16
 matplotlib>=3.9.4
+sphinx-copybutton==0.5.2
 # PyTorch Theme
--e git+https://github.com/pytorch/pytorch_sphinx_theme.git#egg=pytorch_sphinx_theme
-myst-parser==0.18.1
-sphinx_design==0.4.1
-sphinx-copybutton==0.5.0
+-e git+https://github.com/pytorch/pytorch_sphinx_theme.git@pytorch_sphinx_theme2#egg=pytorch_sphinx_theme2
 
 # script unit test requirements
 yaspin==3.1.0

.ci/scripts/build-qnn-sdk.sh

@@ -38,6 +38,7 @@ set_up_aot() {
       -DEXECUTORCH_BUILD_EXTENSION_EXTENSION_LLM=ON \
       -DEXECUTORCH_BUILD_EXTENSION_EXTENSION_LLM_RUNNER=ON \
       -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
+      -DEXECUTORCH_BUILD_EXTENSION_NAMED_DATA_MAP=ON \
       -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
       -DEXECUTORCH_ENABLE_EVENT_TRACER=ON \
       -DPYTHON_EXECUTABLE=python3

.ci/scripts/setup-openvino.sh

@@ -10,19 +10,17 @@ 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$(nproc)
+# Download and install OpenVINO from release packages
+OPENVINO_VERSION="2025.3"
+OPENVINO_BUILD="2025.3.0.19807.44526285f24"
+OPENVINO_URL="https://storage.openvinotoolkit.org/repositories/openvino/packages/${OPENVINO_VERSION}/linux/openvino_toolkit_ubuntu22_${OPENVINO_BUILD}_x86_64.tgz"
 
-cd ..
-cmake --install build --prefix dist
+curl -Lo /tmp/openvino_toolkit.tgz --retry 3 --fail ${OPENVINO_URL}
+tar -xzf /tmp/openvino_toolkit.tgz
+mv openvino_toolkit_ubuntu22_${OPENVINO_BUILD}_x86_64 openvino
 
-source dist/setupvars.sh
-cd ../backends/openvino
+source openvino/setupvars.sh
+cd backends/openvino
 pip install -r requirements.txt
 cd scripts
 ./openvino_build.sh --enable_python

.ci/scripts/setup-samsung-linux-deps.sh

@@ -11,9 +11,9 @@ set -ex
 
 download_ai_lite_core() {
   API_BASE="https://soc-developer.semiconductor.samsung.com/api/v1/resource/ai-litecore/download"
-  API_KEY="kn10SoSY3hkC-9Qny5TqD2mnqVrlupv3krnjLeBt5cY"
+  API_KEY=$SAMSUNG_AI_LITECORE_KEY
 
-  VERSION="0.5"
+  VERSION="0.7"
   OS_NAME="Ubuntu 22.04"
   OUT_FILE="/tmp/exynos-ai-litecore-v${VERSION}.tar.gz"
   TARGET_PATH="/tmp/exynos_ai_lite_core"
@@ -52,7 +52,7 @@ download_ai_lite_core() {
 install_enn_backend() {
   NDK_INSTALLATION_DIR=/opt/ndk
   rm -rf "${NDK_INSTALLATION_DIR}" && sudo mkdir -p "${NDK_INSTALLATION_DIR}"
-  ANDROID_NDK_VERSION=r27b
+  ANDROID_NDK_VERSION=r28c
 
   # build Exynos backend
   export ANDROID_NDK_ROOT=${ANDROID_NDK_ROOT:-/opt/ndk}
@@ -62,7 +62,7 @@ install_enn_backend() {
   export PYTHONPATH=${PYTHONPATH:-}:${EXECUTORCH_ROOT}/..
 }
 
-AI_LITE_CORE_VERSION=0.5.0
+AI_LITE_CORE_VERSION=0.7.0
 
 download_ai_lite_core ${AI_LITE_CORE_VERSION}
 install_enn_backend

.ci/scripts/test-cuda-build.sh

@@ -0,0 +1,95 @@
+#!/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 -exu
+
+CUDA_VERSION=${1:-"12.6"}
+
+echo "=== Testing ExecuTorch CUDA ${CUDA_VERSION} Build ==="
+
+# Function to build and test ExecuTorch with CUDA support
+test_executorch_cuda_build() {
+    local cuda_version=$1
+
+    echo "Building ExecuTorch with CUDA ${cuda_version} support..."
+    echo "ExecuTorch will automatically detect CUDA and install appropriate PyTorch wheel"
+
+    # Check available resources before starting
+    echo "=== System Information ==="
+    echo "Available memory: $(free -h | grep Mem | awk '{print $2}')"
+    echo "Available disk space: $(df -h . | tail -1 | awk '{print $4}')"
+    echo "CPU cores: $(nproc)"
+    echo "CUDA version check:"
+    nvcc --version || echo "nvcc not found"
+    nvidia-smi || echo "nvidia-smi not found"
+
+    # Set CMAKE_ARGS to enable CUDA build - ExecuTorch will handle PyTorch installation automatically
+    export CMAKE_ARGS="-DEXECUTORCH_BUILD_CUDA=ON"
+
+    echo "=== Starting ExecuTorch Installation ==="
+    # Install ExecuTorch with CUDA support with timeout and error handling
+    timeout 5400 ./install_executorch.sh || {
+        local exit_code=$?
+        echo "ERROR: install_executorch.sh failed with exit code: $exit_code"
+        if [ $exit_code -eq 124 ]; then
+            echo "ERROR: Installation timed out after 90 minutes"
+        fi
+        exit $exit_code
+    }
+
+    echo "SUCCESS: ExecuTorch CUDA build completed"
+
+    # Verify the installation
+    echo "=== Verifying ExecuTorch CUDA Installation ==="
+
+    # Test that ExecuTorch was built successfully
+    python -c "
+import executorch
+print('SUCCESS: ExecuTorch imported successfully')
+"
+
+    # Test CUDA availability and show details
+    python -c "
+try:
+    import torch
+    print('INFO: PyTorch version:', torch.__version__)
+    print('INFO: CUDA available:', torch.cuda.is_available())
+
+    if torch.cuda.is_available():
+        print('SUCCESS: CUDA is available for ExecuTorch')
+        print('INFO: CUDA version:', torch.version.cuda)
+        print('INFO: GPU device count:', torch.cuda.device_count())
+        print('INFO: Current GPU device:', torch.cuda.current_device())
+        print('INFO: GPU device name:', torch.cuda.get_device_name())
+
+        # Test basic CUDA tensor operation
+        device = torch.device('cuda')
+        x = torch.randn(10, 10).to(device)
+        y = torch.randn(10, 10).to(device)
+        z = torch.mm(x, y)
+        print('SUCCESS: CUDA tensor operation completed on device:', z.device)
+        print('INFO: Result tensor shape:', z.shape)
+
+        print('SUCCESS: ExecuTorch CUDA integration verified')
+    else:
+        print('WARNING: CUDA not detected, but ExecuTorch built successfully')
+        exit(1)
+except Exception as e:
+    print('ERROR: ExecuTorch CUDA test failed:', e)
+    exit(1)
+"
+
+    echo "SUCCESS: ExecuTorch CUDA ${cuda_version} build and verification completed successfully"
+}
+
+# Main execution
+echo "Current working directory: $(pwd)"
+echo "Directory contents:"
+ls -la
+
+# Run the CUDA build test
+test_executorch_cuda_build "${CUDA_VERSION}"

.ci/scripts/test_backend_linux.sh → .ci/scripts/test_backend.sh

@@ -1,6 +1,7 @@
 #!/usr/bin/env bash
 # Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
+# Copyright 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.
@@ -10,16 +11,26 @@ SUITE=$1
 FLOW=$2
 ARTIFACT_DIR=$3
 
-REPORT_FILE="$ARTIFACT_DIR/test-report-$FLOW-$SUITE.csv"
+REPORT_FILE="$ARTIFACT_DIR/test-report-$FLOW-$SUITE.json"
 
 echo "Running backend test job for suite $SUITE, flow $FLOW."
 echo "Saving job artifacts to $ARTIFACT_DIR."
 
-# The generic Linux job chooses to use base env, not the one setup by the image
 eval "$(conda shell.bash hook)"
 CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
 conda activate "${CONDA_ENV}"
 
+if [[ "$(uname)" == "Darwin" ]]; then
+    bash .ci/scripts/setup-conda.sh
+    eval "$(conda shell.bash hook)"
+    CONDA_RUN_CMD="${CONDA_RUN} --no-capture-output"
+    ${CONDA_RUN_CMD} pip install awscli==1.37.21
+    IS_MACOS=1
+else
+    CONDA_RUN_CMD=""
+    IS_MACOS=0
+fi
+
 export PYTHON_EXECUTABLE=python
 
 # CMake options to use, in addition to the defaults.
@@ -48,13 +59,23 @@ fi
 if [[ "$FLOW" == *arm* ]]; then
     # Setup ARM deps.
     .ci/scripts/setup-arm-baremetal-tools.sh
+    source examples/arm/ethos-u-scratch/setup_path.sh
+
+    if [[ "$FLOW" == *ethos_u* ]]; then
+        # Prepare a test runner binary that can run on the Corstone-3x0 FVPs
+        backends/arm/scripts/build_executorch.sh
+        backends/arm/test/setup_testing.sh
+    fi
 fi
 
-# We need the runner to test the built library.
-PYTHON_EXECUTABLE=python CMAKE_ARGS="$EXTRA_BUILD_ARGS" .ci/scripts/setup-linux.sh --build-tool cmake --build-mode Release --editable true
+if [[ $IS_MACOS -eq 1 ]]; then
+    SETUP_SCRIPT=.ci/scripts/setup-macos.sh
+else
+    SETUP_SCRIPT=.ci/scripts/setup-linux.sh
+fi
+CMAKE_ARGS="$EXTRA_BUILD_ARGS" ${CONDA_RUN_CMD} $SETUP_SCRIPT --build-tool cmake --build-mode Release --editable true
 
 EXIT_CODE=0
-python -m executorch.backends.test.suite.runner $SUITE --flow $FLOW --report "$REPORT_FILE" || EXIT_CODE=$?
-
+${CONDA_RUN_CMD} pytest -c /dev/nul -n auto backends/test/suite/$SUITE/ -m flow_$FLOW --json-report --json-report-file="$REPORT_FILE" || EXIT_CODE=$?
 # Generate markdown summary.
-python -m executorch.backends.test.suite.generate_markdown_summary "$REPORT_FILE" > ${GITHUB_STEP_SUMMARY:-"step_summary.md"} --exit-code $EXIT_CODE
+${CONDA_RUN_CMD} python -m executorch.backends.test.suite.generate_markdown_summary_json "$REPORT_FILE" > ${GITHUB_STEP_SUMMARY:-"step_summary.md"} --exit-code $EXIT_CODE