2025-07-25 nightly release (21c8e67)

Author: pytorchbot

.ci/scripts/unittest-linux.sh

@@ -22,7 +22,7 @@ if [[ "$BUILD_TOOL" == "cmake" ]]; then
 
     # We need the runner to test the built library.
     PYTHON_EXECUTABLE=python \
-    CMAKE_ARGS="-DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
+    CMAKE_ARGS="-DEXECUTORCH_BUILD_EXTENSION_EVALUE_UTIL=ON -DEXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL=ON -DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
     .ci/scripts/setup-linux.sh "$@"
 
     .ci/scripts/unittest-linux-cmake.sh

.ci/scripts/unittest-macos.sh

@@ -22,7 +22,7 @@ trap 'rm -rfv ${TMP_DIR}' EXIT
 # Setup MacOS dependencies as there is no Docker support on MacOS atm
 # We need the runner to test the built library.
 PYTHON_EXECUTABLE=python \
-CMAKE_ARGS="-DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
+CMAKE_ARGS="-DEXECUTORCH_BUILD_EXTENSION_EVALUE_UTIL=ON -DEXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL=ON -DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
 ${CONDA_RUN} --no-capture-output \
 .ci/scripts/setup-macos.sh "$@"
 

.github/workflows/trunk.yml

@@ -269,7 +269,7 @@ jobs:
         if [[ ${{ matrix.os}} == "bare_metal" ]]; then
           bash test/build_size_test.sh "-DCMAKE_TOOLCHAIN_FILE=${toolchain_cmake} -DEXECUTORCH_BUILD_ARM_BAREMETAL=ON"
         elif [[ ${{ matrix.os}} == "zephyr-preset" ]]; then
-          CXXFLAGS=${cxx_flags} cmake --preset zephyr -DCMAKE_BUILD_TYPE=Release -DEXECUTORCH_OPTIMIZE_SIZE=ON -DCMAKE_INSTALL_PREFIX=cmake-out -Bcmake-out .
+          CXXFLAGS=${cxx_flags} cmake --preset zephyr -DCMAKE_BUILD_TYPE=Release -DEXECUTORCH_OPTIMIZE_SIZE=ON -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON -DCMAKE_INSTALL_PREFIX=cmake-out -Bcmake-out .
           cmake --build cmake-out -j9 --target install --config Release
           CXXFLAGS=${cxx_flags}  cmake -DCMAKE_TOOLCHAIN_FILE=${toolchain_cmake} -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=cmake-out -Bcmake-out/test test
           cmake --build cmake-out/test -j9 --config Release
@@ -387,7 +387,7 @@ jobs:
         eval "$(conda shell.bash hook)"
 
         # Install requirements
-        ${CONDA_RUN} python install_executorch.py
+        ${CONDA_RUN} EXECUTORCH_BUILD_TORCHAO=1 python install_executorch.py
         ${CONDA_RUN} sh examples/models/llama/install_requirements.sh
 
         # Run test

CMakeLists.txt

@@ -548,6 +548,16 @@ if(EXECUTORCH_BUILD_EXTENSION_DATA_LOADER)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/data_loader)
 endif()
 
+if(EXECUTORCH_BUILD_EXTENSION_EVALUE_UTIL)
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/evalue_util)
+  install(
+    DIRECTORY extension/evalue_util/
+    DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/executorch/extension/evalue_util
+    FILES_MATCHING
+    PATTERN "*.h"
+  )
+endif()
+
 if(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/flat_tensor)
 endif()
@@ -576,6 +586,12 @@ endif()
 
 if(EXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/runner_util)
+  install(
+    DIRECTORY extension/runner_util/
+    DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/executorch/extension/runner_util
+    FILES_MATCHING
+    PATTERN "*.h"
+  )
 endif()
 
 if(EXECUTORCH_BUILD_EXTENSION_TENSOR)
@@ -651,8 +667,7 @@ if(EXECUTORCH_BUILD_PYBIND)
 
   # util lib
   add_library(
-    util ${CMAKE_CURRENT_SOURCE_DIR}/extension/evalue_util/print_evalue.cpp
-         ${CMAKE_CURRENT_SOURCE_DIR}/extension/aten_util/aten_bridge.cpp
+    util ${CMAKE_CURRENT_SOURCE_DIR}/extension/aten_util/aten_bridge.cpp
   )
   target_include_directories(
     util PUBLIC ${_common_include_directories} ${TORCH_INCLUDE_DIRS}
@@ -695,7 +710,9 @@ endif()
 
 if(EXECUTORCH_BUILD_EXECUTOR_RUNNER)
   # Baseline libraries that executor_runner will link against.
-  set(_executor_runner_libs executorch gflags)
+  set(_executor_runner_libs executorch extension_evalue_util
+                            extension_runner_util gflags
+  )
 
   if(EXECUTORCH_BUILD_KERNELS_OPTIMIZED)
     list(APPEND _executor_runner_libs optimized_native_cpu_ops_lib)

backends/apple/coreml/test/tester.py

@@ -4,23 +4,73 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
-from typing import Any, List, Optional, Tuple
+import functools
+from typing import Any, List, Optional, Sequence, Tuple
 
+import coremltools as ct
 import executorch
 import executorch.backends.test.harness.stages as BaseStages
-
 import torch
+
+from executorch.backends.apple.coreml.compiler import CoreMLBackend
 from executorch.backends.apple.coreml.partition import CoreMLPartitioner
+from executorch.backends.apple.coreml.quantizer import CoreMLQuantizer
 from executorch.backends.test.harness import Tester as TesterBase
 from executorch.backends.test.harness.stages import StageType
 from executorch.exir import EdgeCompileConfig
 from executorch.exir.backend.partitioner import Partitioner
 
 
+def _create_default_partitioner(
+    minimum_deployment_target: Any = ct.target.iOS15,
+) -> CoreMLPartitioner:
+    return CoreMLPartitioner(
+        compile_specs=CoreMLBackend.generate_compile_specs(
+            minimum_deployment_target=minimum_deployment_target
+        )
+    )
+
+
+def _get_static_int8_linear_qconfig():
+    return ct.optimize.torch.quantization.LinearQuantizerConfig(
+        global_config=ct.optimize.torch.quantization.ModuleLinearQuantizerConfig(
+            quantization_scheme="symmetric",
+            activation_dtype=torch.quint8,
+            weight_dtype=torch.qint8,
+            weight_per_channel=True,
+        )
+    )
+
+
+class Quantize(BaseStages.Quantize):
+    def __init__(
+        self,
+        quantizer: Optional[CoreMLQuantizer] = None,
+        quantization_config: Optional[Any] = None,
+        calibrate: bool = True,
+        calibration_samples: Optional[Sequence[Any]] = None,
+        is_qat: Optional[bool] = False,
+    ):
+        super().__init__(
+            quantizer=quantizer
+            or CoreMLQuantizer(
+                quantization_config or _get_static_int8_linear_qconfig()
+            ),
+            calibrate=calibrate,
+            calibration_samples=calibration_samples,
+            is_qat=is_qat,
+        )
+
+
 class Partition(BaseStages.Partition):
-    def __init__(self, partitioner: Optional[Partitioner] = None):
+    def __init__(
+        self,
+        partitioner: Optional[Partitioner] = None,
+        minimum_deployment_target: Optional[Any] = ct.target.iOS15,
+    ):
         super().__init__(
-            partitioner=partitioner or CoreMLPartitioner,
+            partitioner=partitioner
+            or _create_default_partitioner(minimum_deployment_target),
         )
 
 
@@ -29,9 +79,12 @@ def __init__(
         self,
         partitioners: Optional[List[Partitioner]] = None,
         edge_compile_config: Optional[EdgeCompileConfig] = None,
+        minimum_deployment_target: Optional[Any] = ct.target.iOS15,
     ):
         super().__init__(
-            default_partitioner_cls=CoreMLPartitioner,
+            default_partitioner_cls=lambda: _create_default_partitioner(
+                minimum_deployment_target
+            ),
             partitioners=partitioners,
             edge_compile_config=edge_compile_config,
         )
@@ -43,13 +96,20 @@ def __init__(
         module: torch.nn.Module,
         example_inputs: Tuple[torch.Tensor],
         dynamic_shapes: Optional[Tuple[Any]] = None,
+        minimum_deployment_target: Optional[Any] = ct.target.iOS15,
     ):
         # Specialize for XNNPACK
         stage_classes = (
             executorch.backends.test.harness.Tester.default_stage_classes()
             | {
-                StageType.PARTITION: Partition,
-                StageType.TO_EDGE_TRANSFORM_AND_LOWER: ToEdgeTransformAndLower,
+                StageType.QUANTIZE: Quantize,
+                StageType.PARTITION: functools.partial(
+                    Partition, minimum_deployment_target=minimum_deployment_target
+                ),
+                StageType.TO_EDGE_TRANSFORM_AND_LOWER: functools.partial(
+                    ToEdgeTransformAndLower,
+                    minimum_deployment_target=minimum_deployment_target,
+                ),
             }
         )
 

backends/arm/arm_backend.py

@@ -128,7 +128,7 @@ def ethosu_compile_spec(
         self.compiler_flags.append("--output-format=raw")
         self.compiler_flags.append("--debug-force-regor")
 
-        base_tosa_version = "TOSA-0.80+BI"
+        base_tosa_version = "TOSA-1.0+INT"
         if "u55" in target:
             # Add the Ethos-U55 extension marker
             base_tosa_version += "+u55"

backends/arm/quantizer/arm_quantizer_utils.py

@@ -18,22 +18,21 @@
 from torch.fx import GraphModule, Node
 
 from torchao.quantization.pt2e.quantizer import QuantizationAnnotation
+from torchao.quantization.pt2e.quantizer.quantizer import Q_ANNOTATION_KEY
 
 
 def is_annotated(node: Node) -> bool:
     """Given a node return whether the node is annotated."""
     return (
-        "quantization_annotation" in node.meta
-        and cast(
-            QuantizationAnnotation, node.meta["quantization_annotation"]
-        )._annotated
+        Q_ANNOTATION_KEY in node.meta
+        and cast(QuantizationAnnotation, node.meta[Q_ANNOTATION_KEY])._annotated
     )
 
 
 def is_output_annotated(node: Node) -> bool:
     """Given a node, return whether the output of the node is annotated."""
-    if "quantization_annotation" in node.meta:
-        annotation = cast(QuantizationAnnotation, node.meta["quantization_annotation"])
+    if Q_ANNOTATION_KEY in node.meta:
+        annotation = cast(QuantizationAnnotation, node.meta[Q_ANNOTATION_KEY])
         return annotation._annotated and annotation.output_qspec is not None
     else:
         return False
@@ -43,9 +42,9 @@ def mark_node_as_annotated(node: Node) -> None:
     """Marks node as annotated. If needed, an empty  QuantizationAnnotation is added
     to the quantization_annotation node meta entry.
     """
-    if "quantization_annotation" not in node.meta:
-        node.meta["quantization_annotation"] = QuantizationAnnotation()
-    node.meta["quantization_annotation"]._annotated = True
+    if Q_ANNOTATION_KEY not in node.meta:
+        node.meta[Q_ANNOTATION_KEY] = QuantizationAnnotation()
+    node.meta[Q_ANNOTATION_KEY]._annotated = True
 
 
 def is_ok_for_quantization(node: Node, gm: GraphModule):

backends/arm/test/test_arm_baremetal.sh

@@ -228,7 +228,7 @@ test_models_ethos-u85() { # End to End model tests using model_test.py
     python3 backends/arm/test/test_model.py --test_output=arm_test/test_model --target=ethos-u85-256 --model=mv2 --extra_flags="-DET_ATOL=2.00 -DET_RTOL=2.00"
     python3 backends/arm/test/test_model.py --test_output=arm_test/test_model --target=ethos-u85-512 --model=mv3 --extra_flags="-DET_ATOL=5.00 -DET_RTOL=5.00"
     python3 backends/arm/test/test_model.py --test_output=arm_test/test_model --target=ethos-u85-128 --model=lstm --extra_flags="-DET_ATOL=0.03 -DET_RTOL=0.03"
-    python3 backends/arm/test/test_model.py --test_output=arm_test/test_model --target=ethos-u85-128 --model=w2l --extra_flags="-DET_ATOL=0.01 -DET_RTOL=0.01"
+    #python3 backends/arm/test/test_model.py --test_output=arm_test/test_model --target=ethos-u85-128 --model=w2l --extra_flags="-DET_ATOL=0.01 -DET_RTOL=0.01"  # Takes long time to run
     python3 backends/arm/test/test_model.py --test_output=arm_test/test_model --target=ethos-u85-256 --model=ic4 --extra_flags="-DET_ATOL=0.8 -DET_RTOL=0.8" --timeout=2400
     python3 backends/arm/test/test_model.py --test_output=arm_test/test_model --target=ethos-u85-128 --model=resnet18 --extra_flags="-DET_ATOL=0.2 -DET_RTOL=0.2"
     python3 backends/arm/test/test_model.py --test_output=arm_test/test_model --target=ethos-u85-128 --model=resnet50 --extra_flags="-DET_ATOL=0.2 -DET_RTOL=0.2"

backends/cadence/aot/TARGETS

@@ -41,6 +41,7 @@ python_library(
         ":ops_registrations",
         ":passes",
         ":replace_ops",
+        ":compiler_funcs",
         ":utils",
         "//caffe2:torch",
         "//executorch/backends/cadence/aot/quantizer:fusion_pass",
@@ -332,6 +333,18 @@ python_library(
     ],
 )
 
+python_library(
+    name = "compiler_funcs",
+    srcs = [
+        "compiler_funcs.py",
+    ],
+    typing = True,
+    deps = [
+        "//caffe2:torch",
+        "//pytorch/ao:torchao",
+    ],
+)
+
 
 python_unittest(
     name = "test_graph_builder",