Merge remote-tracking branch 'upstream/main'

Author: Jiseong-oh

.ci/scripts/export_model_artifact.sh

@@ -358,7 +358,7 @@ if [ "$MODEL_NAME" = "voxtral_realtime" ]; then
   STREAMING_ARG=""
   PREPROCESSOR_ARGS="--feature_size 128 --output_file ${OUTPUT_DIR}/preprocessor.pte"
   if [ "$USE_STREAMING" = "true" ]; then
-    STREAMING_ARG="--streaming"
+    STREAMING_ARG="--streaming --sliding-window 2048"
     PREPROCESSOR_ARGS="$PREPROCESSOR_ARGS --streaming"
   else
     PREPROCESSOR_ARGS="$PREPROCESSOR_ARGS --stack_output --max_audio_len 300"
@@ -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/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

@@ -165,7 +165,7 @@ jobs:
       contents: read
     steps:
       - name: configure aws credentials
-        uses: aws-actions/configure-aws-credentials@v1.7.0
+        uses: aws-actions/configure-aws-credentials@v4
         with:
           role-to-assume: arn:aws:iam::308535385114:role/gha_executorch_upload-frameworks-android
           aws-region: us-east-1

.github/workflows/apple.yml

@@ -239,7 +239,7 @@ jobs:
           python-version: '3.11'
           cache: pip
       - name: configure aws credentials
-        uses: aws-actions/configure-aws-credentials@v1.7.0
+        uses: aws-actions/configure-aws-credentials@v4
         with:
           role-to-assume: arn:aws:iam::308535385114:role/gha_executorch_upload-frameworks-ios
           aws-region: us-east-1

.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"

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:

backends/arm/requirements-arm-tosa.txt

@@ -9,4 +9,4 @@ tosa-adapter-model-explorer == 0.1.0
 ai-edge-model-explorer >= 0.1.16
 # NOTE: Will be removed when tosa-tools is installed via pypi
 pybind11 == 2.10.4
-pytest-timeout
+pytest-timeout == 2.4.0

backends/arm/test/ops/test_conv2d.py

@@ -487,6 +487,7 @@ def _get_dtype_count(model: torch.nn.Module):
     "test_data",
     test_data_INT,
 )
+@common.SkipIfNoModelConverter
 def test_convolution_2d_vgf_quant_a8w4(test_data):
     model, per_channel_quantization = test_data()
     pipeline = VgfPipeline[input_t](