Merge branch 'main' into add-dtype-check-for-where

YufengShi-dudu · web-flow · commit 62a1cb4d9826 · 2025-09-23T11:06:20.000+01:00
diff --git a/.ci/scripts/setup-samsung-linux-deps.sh b/.ci/scripts/setup-samsung-linux-deps.sh
@@ -11,7 +11,7 @@ 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"
   OS_NAME="Ubuntu 22.04"
diff --git a/.ci/scripts/test-cuda-build.sh b/.ci/scripts/test-cuda-build.sh
@@ -9,14 +9,14 @@ set -exu
 
 CUDA_VERSION=${1:-"12.6"}
 
-echo "=== Testing ExecutorTorch CUDA ${CUDA_VERSION} Build ==="
+echo "=== Testing ExecuTorch CUDA ${CUDA_VERSION} Build ==="
 
-# Function to build and test ExecutorTorch with CUDA support
+# Function to build and test ExecuTorch with CUDA support
 test_executorch_cuda_build() {
     local cuda_version=$1
 
-    echo "Building ExecutorTorch with CUDA ${cuda_version} support..."
-    echo "ExecutorTorch will automatically detect CUDA and install appropriate PyTorch wheel"
+    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 ==="
@@ -27,11 +27,11 @@ test_executorch_cuda_build() {
     nvcc --version || echo "nvcc not found"
     nvidia-smi || echo "nvidia-smi not found"
 
-    # Set CMAKE_ARGS to enable CUDA build - ExecutorTorch will handle PyTorch installation automatically
+    # Set CMAKE_ARGS to enable CUDA build - ExecuTorch will handle PyTorch installation automatically
     export CMAKE_ARGS="-DEXECUTORCH_BUILD_CUDA=ON"
 
-    echo "=== Starting ExecutorTorch Installation ==="
-    # Install ExecutorTorch with CUDA support with timeout and error handling
+    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"
@@ -41,15 +41,15 @@ test_executorch_cuda_build() {
         exit $exit_code
     }
 
-    echo "SUCCESS: ExecutorTorch CUDA build completed"
+    echo "SUCCESS: ExecuTorch CUDA build completed"
 
     # Verify the installation
-    echo "=== Verifying ExecutorTorch CUDA Installation ==="
+    echo "=== Verifying ExecuTorch CUDA Installation ==="
 
-    # Test that ExecutorTorch was built successfully
+    # Test that ExecuTorch was built successfully
     python -c "
 import executorch
-print('SUCCESS: ExecutorTorch imported successfully')
+print('SUCCESS: ExecuTorch imported successfully')
 "
 
     # Test CUDA availability and show details
@@ -60,7 +60,7 @@ try:
     print('INFO: CUDA available:', torch.cuda.is_available())
 
     if torch.cuda.is_available():
-        print('SUCCESS: CUDA is available for ExecutorTorch')
+        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())
@@ -74,16 +74,16 @@ try:
         print('SUCCESS: CUDA tensor operation completed on device:', z.device)
         print('INFO: Result tensor shape:', z.shape)
 
-        print('SUCCESS: ExecutorTorch CUDA integration verified')
+        print('SUCCESS: ExecuTorch CUDA integration verified')
     else:
-        print('WARNING: CUDA not detected, but ExecutorTorch built successfully')
+        print('WARNING: CUDA not detected, but ExecuTorch built successfully')
         exit(1)
 except Exception as e:
-    print('ERROR: ExecutorTorch CUDA test failed:', e)
+    print('ERROR: ExecuTorch CUDA test failed:', e)
     exit(1)
 "
 
-    echo "SUCCESS: ExecutorTorch CUDA ${cuda_version} build and verification completed successfully"
+    echo "SUCCESS: ExecuTorch CUDA ${cuda_version} build and verification completed successfully"
 }
 
 # Main execution
diff --git a/.github/scripts/propose_ghstack_orig_pr.py b/.github/scripts/propose_ghstack_orig_pr.py
@@ -86,6 +86,17 @@ def get_pr_stack_from_number(ref: str, repo: Repository) -> List[int]:
     return pr_stack
 
 
+def get_differential_revision(pr, repo: Repository) -> str:
+    body = repo.get_pull(pr.number).body
+    matches = re.findall(r"Differential Revision: .*", body)
+    count = len(matches)
+    if count == 1:
+        # If there's more than one Differential Revision, let's just return empty
+        # so that we can disambiguate manually.
+        return matches[0]
+    return ""
+
+
 def create_prs_for_orig_branch(pr_stack: List[int], repo: Repository):
     # For the first PR, we want to merge to `main` branch, and we will update
     # as we go through the stack
@@ -100,13 +111,15 @@ def create_prs_for_orig_branch(pr_stack: List[int], repo: Repository):
         # The PR we want to create is then "branch_to_merge" <- gh/user/x/orig
         # gh/user/x/orig is the clean diff between gh/user/x/base <- gh/user/x/head
         orig_branch_merge_head = pr.base.ref.replace("base", "orig")
+        differential_revision_text = get_differential_revision(pr, repo)
         bot_metadata = f"""This PR was created by the merge bot to help merge the original PR into the main branch.
 ghstack PR number: https://github.com/pytorch/executorch/pull/{pr.number} by @{pr.user.login}
 ^ Please use this as the source of truth for the PR details, comments, and reviews
 ghstack PR base: https://github.com/pytorch/executorch/tree/{pr.base.ref}
 ghstack PR head: https://github.com/pytorch/executorch/tree/{pr.head.ref}
 Merge bot PR base: https://github.com/pytorch/executorch/tree/{orig_branch_merge_base}
 Merge bot PR head: https://github.com/pytorch/executorch/tree/{orig_branch_merge_head}
+{differential_revision_text}
 @diff-train-skip-merge"""
 
         existing_orig_pr = repo.get_pulls(
diff --git a/.github/workflows/pull.yml b/.github/workflows/pull.yml
@@ -900,12 +900,14 @@ jobs:
     permissions:
       id-token: write
       contents: read
+    secrets: inherit
     with:
       runner: linux.2xlarge
       docker-image: ci-image:executorch-ubuntu-22.04-clang12-android
       submodules: 'recursive'
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
       timeout: 90
+      secrets-env: SAMSUNG_AI_LITECORE_KEY
       script: |
         set -ex
 
@@ -917,6 +919,7 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh --build-tool "cmake"
 
         # Setup Samsung SDK (AI Lite Core) and install enn backend
+        export SAMSUNG_AI_LITECORE_KEY=$SECRET_SAMSUNG_AI_LITECORE_KEY
         source .ci/scripts/setup-samsung-linux-deps.sh
 
         # Test models serially
diff --git a/.github/workflows/test-cuda-builds.yml b/.github/workflows/test-cuda-builds.yml
@@ -24,7 +24,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        cuda-version: ["12.6", "12.8", "12.9"]
+        cuda-version: ["12.6", "12.8", "13.0"]
 
     name: test-executorch-cuda-build-${{ matrix.cuda-version }}
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
@@ -199,8 +199,7 @@ We use [`lintrunner`](https://pypi.org/project/lintrunner/) to help make sure th
 code follows our standards. Set it up with:
 
 ```
-pip install lintrunner==0.12.7
-pip install lintrunner-adapters==0.12.4
+./install_requirements.sh  # (automatically run by install_executorch.sh)
 lintrunner init
 ```
 
diff --git a/backends/cortex_m/test/test_quantize_op_fusion_pass.py b/backends/cortex_m/test/test_quantize_op_fusion_pass.py
@@ -313,7 +313,7 @@ def forward(self, x, y):
         # Apply passes
         transformed_program = self._apply_passes(edge_program)
 
-        # Generate ExecutorTorch program
+        # Generate ExecuTorch program
         executorch_program = transformed_program.to_executorch()
 
         # Verify the program contains the expected fused operator
diff --git a/backends/vulkan/test/utils.py b/backends/vulkan/test/utils.py
@@ -303,13 +303,13 @@ def run_and_check_output(
     Returns:
         bool: True if outputs match within tolerance, False otherwise
     """
-    # Load the ExecutorTorch program
+    # Load the ExecuTorch program
     executorch_module = _load_for_executorch_from_buffer(executorch_program.buffer)
 
     # Flatten inputs for execution
     inputs_flattened, _ = tree_flatten(sample_inputs)
 
-    # Run the ExecutorTorch program
+    # Run the ExecuTorch program
     model_output = executorch_module.run_method("forward", tuple(inputs_flattened))
 
     # Generate reference outputs using the reference model
diff --git a/backends/xnnpack/runtime/XNNPACKBackend.h b/backends/xnnpack/runtime/XNNPACKBackend.h
@@ -1,7 +1,5 @@
 #pragma once
 
-#include <executorch/runtime/platform/compiler.h>
-
 namespace executorch::backends::xnnpack {
 /// The key for the backend. This is used to register the backend, check
 /// availability, and get/set options.
diff --git a/backends/xnnpack/targets.bzl b/backends/xnnpack/targets.bzl
@@ -73,3 +73,13 @@ def define_common_targets():
             # @lint-ignore BUCKLINT: Avoid `link_whole=True` (https://fburl.com/avoid-link-whole)
             link_whole = True,
         )
+    
+    runtime.cxx_library(
+        name = "xnnpack_interface",
+        visibility = [
+            "@EXECUTORCH_CLIENTS",
+        ],
+        exported_headers = [
+            "runtime/XNNPACKBackend.h",
+        ],
+    )
diff --git a/devtools/scripts/profile_model.sh b/devtools/scripts/profile_model.sh
@@ -7,7 +7,7 @@
 
 #!/bin/bash
 
-# ExecutorTorch Model Profiling Script
+# ExecuTorch Model Profiling Script
 #
 # This script automates the process of building executor_runner with profiling enabled,
 # running model inference with ETDump collection, and generating CSV profiling reports.
diff --git a/docs/source/llm/export-llm.md b/docs/source/llm/export-llm.md
@@ -4,7 +4,7 @@ Instead of needing to manually write code to call torch.export(), use ExecuTorch
 
 ## Prerequisites
 
-The LLM export functionality requires the `pytorch_tokenizers` package. If you encounter a `ModuleNotFoundError: No module named 'pytorch_tokenizers'` error, install it from the ExecutorTorch source code:
+The LLM export functionality requires the `pytorch_tokenizers` package. If you encounter a `ModuleNotFoundError: No module named 'pytorch_tokenizers'` error, install it from the ExecuTorch source code:
 
 ```bash
 pip install -e ./extension/llm/tokenizers/
diff --git a/docs/source/ptd-file-format.md b/docs/source/ptd-file-format.md
@@ -111,7 +111,7 @@ The flatbuffer-encoded metadata follows the headers and contains:
 ### Tensor Layout
 
 If a data segment contains a canonical tensor, it may have associated layout information:
-- **Scalar type**: Data type (float32, int32, etc.) using ExecutorTorch scalar types.
+- **Scalar type**: Data type (float32, int32, etc.) using ExecuTorch scalar types.
 - **Sizes**: Dimensions of the tensor.
 - **Dim order**: Memory layout order specifying how dimensions are arranged in memory.
 
diff --git a/docs/source/using-executorch-faqs.md b/docs/source/using-executorch-faqs.md
@@ -16,7 +16,7 @@ if you are using Ubuntu, or use an equivalent install command.
 
 ### ModuleNotFoundError: No module named 'pytorch_tokenizers'
 
-The `pytorch_tokenizers` package is required for LLM export functionality. Install it from the ExecutorTorch source code:
+The `pytorch_tokenizers` package is required for LLM export functionality. Install it from the ExecuTorch source code:
 ```
 pip install -e ./extension/llm/tokenizers/
 ```
diff --git a/examples/vulkan/export.py b/examples/vulkan/export.py
@@ -210,7 +210,7 @@ def main() -> None:
         else:
             logging.error("✗ Model test FAILED - outputs do not match reference")
             raise RuntimeError(
-                "Model validation failed: ExecutorTorch outputs do not match reference model outputs"
+                "Model validation failed: ExecuTorch outputs do not match reference model outputs"
             )
 
     if args.bundled:
diff --git a/exir/emit/test/test_emit.py b/exir/emit/test/test_emit.py
@@ -1920,7 +1920,7 @@ def forward(self, x):
             program_buffer = et_program.buffer
             et_module = _load_for_executorch_from_buffer(program_buffer)
             for _, (inp, expected) in enumerate(zip(test_inputs, reference_outputs)):
-                # Execute with ExecutorTorch
+                # Execute with ExecuTorch
                 et_output = et_module.forward([inp])
                 et_result = et_output[0]  # Get first output
                 # Compare results
diff --git a/exir/passes/memory_planning_pass.py b/exir/passes/memory_planning_pass.py
@@ -287,7 +287,7 @@ def run(
         return PassResult(graph_module, True)
 
     def run_multimethod(self):
-        "Resolve any memory planning done across entry points"
+        """Resolve any memory planning done across entry points, called after run is called on all entry points."""
         if self.share_mutable_buffers:
             arena: int = 0
 
diff --git a/extension/android/executorch_android/src/main/java/org/pytorch/executorch/ExecutorchRuntimeException.java b/extension/android/executorch_android/src/main/java/org/pytorch/executorch/ExecutorchRuntimeException.java
@@ -75,6 +75,7 @@ public class ExecutorchRuntimeException extends RuntimeException {
   }
 
   static class ErrorHelper {
+    private static final boolean ENABLE_READ_LOG_BUFFER = false;
     // Reusable StringBuilder instance
     private static final StringBuilder sb = new StringBuilder();
 
@@ -92,14 +93,15 @@ static String formatMessage(int errorCode, String details) {
             .append("] ")
             .append(baseMessage)
             .append(": ")
-            .append(details)
-            .append("\nDetailed Logs:\n");
-
-        try {
-          String[] logEntries = readLogBuffer(); // JNI call
-          formatLogEntries(sb, logEntries);
-        } catch (Exception e) {
-          sb.append("Failed to retrieve detailed logs: ").append(e.getMessage());
+            .append(details);
+        if (ENABLE_READ_LOG_BUFFER) {
+          try {
+            sb.append("\nDetailed Logs:\n");
+            String[] logEntries = readLogBuffer(); // JNI call
+            formatLogEntries(sb, logEntries);
+          } catch (Exception e) {
+            sb.append("Failed to retrieve detailed logs: ").append(e.getMessage());
+          }
         }
 
         return sb.toString();
diff --git a/extension/android/executorch_android/src/main/java/org/pytorch/executorch/Module.java b/extension/android/executorch_android/src/main/java/org/pytorch/executorch/Module.java
@@ -204,10 +204,9 @@ public MethodMetadata getMethodMetadata(String name) {
       throw new RuntimeException("method " + name + "does not exist for this module");
     }
 
-    MethodMetadata methodMetadata =mMethodMetadata.get(name);
+    MethodMetadata methodMetadata = mMethodMetadata.get(name);
     if (methodMetadata != null) {
       methodMetadata.setBackends(getUsedBackends(name));
-
     }
     return methodMetadata;
   }
diff --git a/extension/android/jni/jni_helper.cpp b/extension/android/jni/jni_helper.cpp
@@ -19,11 +19,12 @@ void throwExecutorchException(uint32_t errorCode, const std::string& details) {
       "org/pytorch/executorch/ExecutorchRuntimeException");
 
   // Find the static factory method: makeExecutorchException(int, String)
-  static auto makeExceptionMethod = exceptionClass->getStaticMethod<
-      facebook::jni::local_ref<facebook::jni::JThrowable>(
-          int, facebook::jni::alias_ref<facebook::jni::JString>)>(
-      "makeExecutorchException",
-      "(ILjava/lang/String;)Lorg/pytorch/executorch/ExecutorchRuntimeException;");
+  static auto makeExceptionMethod =
+      exceptionClass
+          ->getStaticMethod<facebook::jni::local_ref<facebook::jni::JThrowable>(
+              int, facebook::jni::alias_ref<facebook::jni::JString>)>(
+              "makeExecutorchException",
+              "(ILjava/lang/String;)Ljava/lang/RuntimeException;");
 
   auto jDetails = facebook::jni::make_jstring(details);
   // Call the factory method to create the exception object
diff --git a/extension/audio/mel_spectrogram.py b/extension/audio/mel_spectrogram.py
@@ -213,7 +213,7 @@ def export_processor(model=None, output_file="whisper_preprocess.pte"):
 
 def main():
     parser = argparse.ArgumentParser(
-        description="Export WhisperAudioProcessor to ExecutorTorch"
+        description="Export WhisperAudioProcessor to ExecuTorch"
     )
     parser.add_argument(
         "--feature_size",
diff --git a/extension/llm/runner/README.md b/extension/llm/runner/README.md
@@ -1,6 +1,6 @@
-# LLM Runner Framework for ExecutorTorch
+# LLM Runner Framework for ExecuTorch
 
-This directory contains the LLM Runner framework for ExecutorTorch, providing high-level C++ APIs for running Large Language Models with both text-only and multimodal capabilities.
+This directory contains the LLM Runner framework for ExecuTorch, providing high-level C++ APIs for running Large Language Models with both text-only and multimodal capabilities.
 
 ## Overview
 
@@ -474,7 +474,7 @@ DeepFusion is another popular model architecture type where a pretrained encoder
 - **Llama 3.2 Vision**: Uses cross-attention layers for vision-text fusion
 - **Other cross-attention based multimodal models**
 
-For DeepFusion support, consider using the model's native inference framework or wait for future ExecutorTorch updates that may include DeepFusion architecture support.
+For DeepFusion support, consider using the model's native inference framework or wait for future ExecuTorch updates that may include DeepFusion architecture support.
 
 ## Building and Integration
 
diff --git a/extension/llm/runner/multimodal_runner.cpp b/extension/llm/runner/multimodal_runner.cpp
@@ -62,7 +62,7 @@ Error MultimodalRunner::load() {
     ET_LOG(Info, format, __VA_ARGS__);     \
   }
 
-Error MultimodalRunner::prefill(std::vector<MultimodalInput>& inputs) {
+Error MultimodalRunner::prefill(const std::vector<MultimodalInput>& inputs) {
   if (!is_loaded()) {
     ET_CHECK_OK_OR_RETURN_ERROR(load());
   }
diff --git a/extension/llm/runner/multimodal_runner.h b/extension/llm/runner/multimodal_runner.h
@@ -126,7 +126,7 @@ class ET_EXPERIMENTAL MultimodalRunner {
    * @return The error code. KV cache position is tracked internally in pos_.
    */
   virtual ::executorch::runtime::Error prefill(
-      std::vector<MultimodalInput>& inputs);
+      const std::vector<MultimodalInput>& inputs);
 
   inline void stop() {
     text_token_generator_->stop();
diff --git a/extension/module/test/targets.bzl b/extension/module/test/targets.bzl
@@ -19,6 +19,7 @@ def define_common_targets(is_fbcode=False):
             "ET_MODULE_ADD_PATH": "$(location fbcode//executorch/test/models:exported_programs[ModuleAdd.pte])",
             "ET_MODULE_ADD_MUL_PROGRAM_PATH": "$(location fbcode//executorch/test/models:exported_program_and_data[ModuleAddMul.pte])",
             "ET_MODULE_ADD_MUL_DATA_PATH": "$(location fbcode//executorch/test/models:exported_program_and_data[ModuleAddMul.ptd])",
+            "ET_MODULE_SHARED_STATE": "$(location fbcode//executorch/test/models:exported_programs[ModuleSharedState.pte])",
         }
 
         for aten_mode in get_aten_mode_options():
diff --git a/extension/pybindings/portable_lib.py b/extension/pybindings/portable_lib.py
diff --git a/extension/training/examples/CIFAR/data_utils.py b/extension/training/examples/CIFAR/data_utils.py
diff --git a/extension/training/examples/CIFAR/export.py b/extension/training/examples/CIFAR/export.py
diff --git a/extension/training/examples/CIFAR/main.py b/extension/training/examples/CIFAR/main.py
diff --git a/extension/training/examples/CIFAR/train_utils.py b/extension/training/examples/CIFAR/train_utils.py
diff --git a/install_requirements.py b/install_requirements.py
diff --git a/runtime/kernel/targets.bzl b/runtime/kernel/targets.bzl
diff --git a/test/end2end/exported_module.py b/test/end2end/exported_module.py
diff --git a/test/models/export_program.py b/test/models/export_program.py
diff --git a/test/models/targets.bzl b/test/models/targets.bzl
