Merge branch 'main' into export-D97662639

Author: alexey-sidnev

.ci/scripts/export_model_artifact.sh

@@ -424,6 +424,7 @@ if [ "$MODEL_NAME" = "qwen3_5_moe" ]; then
   test -f "${OUTPUT_DIR}/model.pte"
   test -f "${OUTPUT_DIR}/aoti_cuda_blob.ptd"
   ls -al "${OUTPUT_DIR}"
+
   exit 0
 fi
 

.ci/scripts/test_model_e2e_windows.ps1

@@ -135,7 +135,19 @@ try {
     Write-Host "::group::Check CUDA toolchain"
     $nvccOutput = nvcc --version | Out-String
     Write-Host $nvccOutput
-    nvidia-smi
+    $nvidiaSmiCmd = Get-Command nvidia-smi -ErrorAction SilentlyContinue
+    if ($null -eq $nvidiaSmiCmd) {
+        Write-Host "nvidia-smi not available (command not found; driver may not be installed)"
+    }
+    else {
+        try {
+            nvidia-smi
+        }
+        catch {
+            Write-Host "nvidia-smi failed (driver or GPU issue). Error details:"
+            Write-Host $_
+        }
+    }
     if (-not [string]::IsNullOrWhiteSpace($ExpectedCudaVersion)) {
         $versionMatch = [Regex]::Match($nvccOutput, "release\s+(\d+\.\d+)")
         if (-not $versionMatch.Success) {

.ci/scripts/wheel/test_linux.py

@@ -11,18 +11,25 @@
 from examples.models import Backend, Model
 
 if __name__ == "__main__":
-    # On Linux x86_64 the wheel is built with the Qualcomm backend.
-    # Verify that it was registered correctly.
-    if platform.system() == "Linux" and platform.machine() in ("x86_64", "amd64"):
+    if platform.system() == "Linux":
         from executorch.extension.pybindings.portable_lib import (
             _get_registered_backend_names,
         )
 
         registered = _get_registered_backend_names()
+
+        # QNN backend is only available on x86_64.
+        if platform.machine() in ("x86_64", "amd64"):
+            assert (
+                "QnnBackend" in registered
+            ), f"QnnBackend not found in registered backends: {registered}"
+            print("✓ QnnBackend is registered")
+
+        # OpenVINO backend is available on all Linux architectures.
         assert (
-            "QnnBackend" in registered
-        ), f"QnnBackend not found in registered backends: {registered}"
-        print("✓ QnnBackend is registered")
+            "OpenvinoBackend" in registered
+        ), f"OpenvinoBackend not found in registered backends: {registered}"
+        print("✓ OpenvinoBackend is registered")
 
     test_base.run_tests(
         model_tests=[

.ci/scripts/wheel/test_linux_aarch64.py

@@ -12,6 +12,20 @@
     # coremltools does not support linux aarch64 yet and install from the source fails on runtime
     # https://github.com/apple/coremltools/issues/1254
     # https://github.com/apple/coremltools/issues/2195
+
+    from executorch.extension.pybindings.portable_lib import (
+        _get_registered_backend_names,
+    )
+
+    registered = _get_registered_backend_names()
+
+    # OpenVINO backend uses dlopen (no build-time SDK dependency), so it
+    # is compiled into the wheel on all Linux architectures.
+    assert (
+        "OpenvinoBackend" in registered
+    ), f"OpenvinoBackend not found in registered backends: {registered}"
+    print("✓ OpenvinoBackend is registered")
+
     test_base.run_tests(
         model_tests=[
             test_base.ModelTest(

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

@@ -145,9 +145,9 @@ jobs:
         export BUILD_AAR_DIR=aar-out
         bash scripts/build_android_library.sh
         mkdir -p "${ARTIFACTS_DIR_NAME}"
-        cp aar-out/executorch.aar "${ARTIFACTS_DIR_NAME}/executorch.aar"
+        cp aar-out/executorch.aar "${ARTIFACTS_DIR_NAME}/executorch-${FLAVOR}.aar"
 
-        shasum -a 256 "${ARTIFACTS_DIR_NAME}/executorch.aar"
+        shasum -a 256 "${ARTIFACTS_DIR_NAME}/executorch-${FLAVOR}.aar"
 
         # Publish to maven staging
         UPLOAD_TO_MAVEN="${{ inputs.upload_to_maven }}"
@@ -172,11 +172,6 @@ jobs:
       - name: Upload AAR RC to AWS S3
         shell: bash
         run: |
-          wget https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/executorch.aar
-          shasum -a 256 executorch.aar > executorch.aar.sha256sums
-
-          pip install awscli==1.32.18
-          AWS_CMD="aws s3 cp"
           VERSION="${{ inputs.version }}"
           FLAVOR="${{ inputs.flavor }}"
           if [ -z "$VERSION" ]; then
@@ -185,5 +180,11 @@ jobs:
           if [ -z "$FLAVOR" ]; then
             FLAVOR="xnnpack"
           fi
+          wget https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/executorch-${FLAVOR}.aar
+          mv executorch-${FLAVOR}.aar executorch.aar
+          shasum -a 256 executorch.aar > executorch.aar.sha256sums
+
+          pip install awscli==1.32.18
+          AWS_CMD="aws s3 cp"
           ${AWS_CMD} executorch.aar s3://ossci-android/executorch/release/${VERSION}-${FLAVOR}/executorch.aar --acl public-read
           ${AWS_CMD} executorch.aar.sha256sums s3://ossci-android/executorch/release/${VERSION}-${FLAVOR}/executorch.aar.sha256sums --acl public-read

.github/workflows/android-release-on-tag.yml

@@ -6,6 +6,10 @@ on:
       - 'v*.*.*-rc*'
       - 'v*.*.*'
 
+permissions:
+  id-token: write
+  contents: read
+
 jobs:
   prepare:
     runs-on: ubuntu-latest

.github/workflows/cuda-windows.yml

@@ -48,9 +48,11 @@ jobs:
           - model_repo: "nvidia"
             model_name: "parakeet-tdt"
             quant: "quantized-int4-weight-only"
-          - model_repo: "nvidia"
-            model_name: "diar_streaming_sortformer_4spk-v2"
-            quant: "non-quantized"
+          # TODO: sortformer produces 0 segments on Windows after D97788666.
+          # Temporarily disabled until root cause is debugged.
+          # - model_repo: "nvidia"
+          #   model_name: "diar_streaming_sortformer_4spk-v2"
+          #   quant: "non-quantized"
           - model_repo: "mistralai"
             model_name: "Voxtral-Mini-4B-Realtime-2602"
             quant: "quantized-int4-tile-packed"
@@ -129,9 +131,11 @@ jobs:
           - model_repo: "nvidia"
             model_name: "parakeet-tdt"
             quant: "quantized-int4-weight-only"
-          - model_repo: "nvidia"
-            model_name: "diar_streaming_sortformer_4spk-v2"
-            quant: "non-quantized"
+          # TODO: sortformer produces 0 segments on Windows after D97788666.
+          # Temporarily disabled until root cause is debugged.
+          # - model_repo: "nvidia"
+          #   model_name: "diar_streaming_sortformer_4spk-v2"
+          #   quant: "non-quantized"
           - model_repo: "mistralai"
             model_name: "Voxtral-Mini-4B-Realtime-2602"
             quant: "quantized-int4-tile-packed"

CMakeLists.txt

@@ -846,6 +846,9 @@ if(EXECUTORCH_BUILD_KERNELS_TORCHAO)
   set(TORCHAO_ENABLE_ARM_NEON_DOT ON)
   set(TORCHAO_BUILD_KLEIDIAI ON)
 
+  # TorchAO requires find_package(Torch) for include dirs; add its cmake path.
+  add_torch_to_cmake_prefix_path()
+
   # TorchAO kernels look for EXECUTORCH_INCLUDE_DIRS
   if(DEFINED EXECUTORCH_INCLUDE_DIRS)
     message(FATAL_ERROR "EXECUTORCH_INCLUDE_DIRS is already defined")

Makefile

@@ -296,18 +296,21 @@ voxtral_realtime-cuda:
 	@echo "  Binary: cmake-out/examples/models/voxtral_realtime/voxtral_realtime_runner"
 
 silero-vad-cpu:
-	@echo "==> Building and installing ExecuTorch..."
-	cmake --workflow --preset llm-release
-	@echo "==> Building Silero VAD runner (CPU)..."
+	@echo "==> Configuring and installing ExecuTorch (without LLM runner)..."
+	cmake --preset llm-release -DEXECUTORCH_BUILD_EXTENSION_LLM_RUNNER=OFF
+	cmake --build cmake-out --parallel "$$(sysctl -n hw.ncpu)"
+	cmake --install cmake-out
+	@echo "==> Building Silero VAD runners (CPU)..."
 	cmake -DCMAKE_BUILD_TYPE=Release \
 		-DCMAKE_FIND_ROOT_PATH=$(CURDIR)/cmake-out \
 		-DCMAKE_PREFIX_PATH=$(CURDIR)/cmake-out \
 		-S examples/models/silero_vad \
 		-B cmake-out/examples/models/silero_vad
-	cmake --build cmake-out/examples/models/silero_vad --target silero_vad_runner
+	cmake --build cmake-out/examples/models/silero_vad --target silero_vad_runner silero_vad_stream_runner
 	@echo ""
 	@echo "✓ Build complete!"
 	@echo "  Binary: cmake-out/examples/models/silero_vad/silero_vad_runner"
+	@echo "  Binary: cmake-out/examples/models/silero_vad/silero_vad_stream_runner"
 
 llama-cpu:
 	@echo "==> Building and installing ExecuTorch..."

README-wheel.md

@@ -6,14 +6,18 @@ PyTorch programs.
 
 The `executorch` pip package is in beta.
 * Supported python versions: 3.10, 3.11, 3.12, 3.13
-* Compatible systems: Linux x86_64, macOS aarch64
+* Compatible systems: Linux x86_64, Linux aarch64, macOS aarch64
 
 The prebuilt `executorch.runtime` module included in this package provides a way
 to run ExecuTorch `.pte` files, with some restrictions:
 * Only [core ATen operators](docs/source/ir-ops-set-definition.md) are linked into the prebuilt module
 * Only the [XNNPACK backend delegate](docs/source/backends/xnnpack/xnnpack-overview.md) is linked into the prebuilt module.
 * \[macOS only] [Core ML](docs/source/backends/coreml/coreml-overview.md) and [MPS](docs/source/backends/mps/mps-overview.md) backend
   are also linked into the prebuilt module.
+* \[Linux x86_64] [QNN](docs/source/backends-qualcomm.md) backend is linked into the prebuilt module.
+* \[Linux] [OpenVINO](docs/source/build-run-openvino.md) backend is also linked into the
+  prebuilt module. OpenVINO requires the runtime to be installed separately:
+  `pip install executorch[openvino]`
 
 Please visit the [ExecuTorch website](https://pytorch.org/executorch) for
 tutorials and documentation. Here are some starting points: