Merge branch 'main' into check-tensor-dim-in-IR-verifier

Author: patelvndn

.ci/docker/requirements-ci.txt

@@ -17,6 +17,7 @@ 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-reredirects==0.1.4
 sphinx-gallery==0.14.0
 breathe==4.34.0
 exhale==0.2.3

.ci/scripts/gather_benchmark_configs.py

@@ -23,6 +23,7 @@
     "samsung_galaxy_s22": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/e59f866a-30aa-4aa1-87b7-4510e5820dfa",
     "samsung_galaxy_s24": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/98f8788c-2e25-4a3c-8bb2-0d1e8897c0db",
     "google_pixel_8_pro": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/d65096ab-900b-4521-be8b-a3619b69236a",
+    "google_pixel_3_private_rooted": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/98d23ca8-ea9e-4fb7-b725-d402017b198d",
 }
 
 # Predefined benchmark configurations

.github/workflows/_android.yml

@@ -22,6 +22,10 @@ jobs:
       script: |
         set -eux
 
+        # Use sccache for NDK compiler as well
+        export CMAKE_CXX_COMPILER_LAUNCHER=sccache
+        export CMAKE_C_COMPILER_LAUNCHER=sccache
+
         # The generic Linux job chooses to use base env, not the one setup by the image
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"
@@ -131,7 +135,8 @@ jobs:
           # https://github.com/ReactiveCircus/android-emulator-runner. The max number
           # of cores we can set is 6, any higher number will be reduced to 6.
           cores: 6
-          ram-size: 12288M
+          ram-size: 16384M
+          heap-size: 12288M
           force-avd-creation: false
           disable-animations: true
           emulator-options: -no-snapshot-save -no-window -gpu swiftshader_indirect -noaudio -no-boot-anim -camera-back none

.github/workflows/android-perf-private-device-experiment.yml

@@ -0,0 +1,62 @@
+name: android-perf (private devices)
+
+on:
+  schedule:
+    - cron: 0 0,4,8,12,16,20 * * *
+  pull_request:
+    paths:
+      - .github/workflows/android-perf-private-device-experiment.yml
+  push:
+    branches:
+      - main
+    paths:
+      - .github/workflows/android-perf-private-device-experiment.yml
+  # Note: GitHub has an upper limit of 10 inputs
+  workflow_dispatch:
+    inputs:
+      models:
+        description: Models to be benchmarked
+        required: false
+        type: string
+        default: mv3,meta-llama/Llama-3.2-1B-Instruct-SpinQuant_INT4_EO8,meta-llama/Llama-3.2-1B-Instruct-QLORA_INT4_EO8
+      devices:
+        description: Target devices to run benchmark
+        required: false
+        type: string
+        default: google_pixel_3_private_rooted
+      benchmark_configs:
+        description: The list of configs used the benchmark
+        required: false
+        type: string
+  workflow_call:
+    inputs:
+      models:
+        description: Models to be benchmarked
+        required: false
+        type: string
+        default: mv3,meta-llama/Llama-3.2-1B-Instruct-SpinQuant_INT4_EO8,meta-llama/Llama-3.2-1B-Instruct-QLORA_INT4_EO8
+      devices:
+        description: Target devices to run benchmark
+        required: false
+        type: string
+        default: google_pixel_3_private_rooted
+      benchmark_configs:
+        description: The list of configs used the benchmark
+        required: false
+        type: string
+
+concurrency:
+  group: android-perf-private-devices-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}
+  cancel-in-progress: true
+
+jobs:
+  android:
+    uses: ./.github/workflows/android-perf.yml
+    secrets: inherit
+    permissions:
+      id-token: write
+      contents: read
+    with:
+      models: ${{ inputs.models || 'mv3,meta-llama/Llama-3.2-1B-Instruct-SpinQuant_INT4_EO8,meta-llama/Llama-3.2-1B-Instruct-QLORA_INT4_EO8' }}
+      devices: google_pixel_3_private_rooted
+      benchmark_configs: ${{ inputs.benchmark_configs }}

.github/workflows/android-perf.yml

@@ -353,6 +353,10 @@ jobs:
       script: |
         set -eux
 
+        # Use sccache for NDK compiler as well
+        export CMAKE_CXX_COMPILER_LAUNCHER=sccache
+        export CMAKE_C_COMPILER_LAUNCHER=sccache
+
         # The generic Linux job chooses to use base env, not the one setup by the image
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"
@@ -392,7 +396,7 @@ jobs:
       fail-fast: false
     with:
       # Due to scheduling a job may be pushed beyond the default 60m threshold
-      timeout: 120
+      timeout: 240
       device-type: android
       runner: linux.2xlarge
       test-infra-ref: ''

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

@@ -49,7 +49,8 @@ jobs:
       contents: read
     with:
       secrets-env: EXECUTORCH_MAVEN_SIGNING_KEYID EXECUTORCH_MAVEN_SIGNING_PASSWORD EXECUTORCH_MAVEN_CENTRAL_PASSWORD EXECUTORCH_MAVEN_CENTRAL_USERNAME EXECUTORCH_MAVEN_SIGNING_GPG_KEY_CONTENTS
-      runner: linux.2xlarge
+      # As this job has access to Maven credential, run this on a fresh ephemeral runner
+      runner: ephemeral.linux.2xlarge
       docker-image: executorch-ubuntu-22.04-clang12-android
       submodules: 'recursive'
       ref: ${{ github.sha }}
@@ -59,6 +60,10 @@ jobs:
       script: |
         set -eux
 
+        # Use sccache for NDK compiler as well
+        export CMAKE_CXX_COMPILER_LAUNCHER=sccache
+        export CMAKE_C_COMPILER_LAUNCHER=sccache
+
         # The generic Linux job chooses to use base env, not the one setup by the image
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"
@@ -75,6 +80,12 @@ jobs:
 
         echo -n "$SECRET_EXECUTORCH_MAVEN_SIGNING_GPG_KEY_CONTENTS" | base64 -d > /tmp/secring.gpg
 
+        # Update the version name in build.gradle in case of maven publish
+        VERSION="${{ inputs.version }}"
+        if [ ! -z "$VERSION" ]; then
+          sed -i "s/\(coordinates(\"org.pytorch\", \"executorch-android\", \"\)\([0-9]\+.[0-9]\+.[0-9]\+\)\(\")\)/\1$VERSION\3/" extension/android/executorch_android/build.gradle
+        fi
+
         # Build AAR Package
         mkdir aar-out
         export BUILD_AAR_DIR=aar-out
@@ -87,7 +98,7 @@ jobs:
         # Publish to maven staging
         UPLOAD_TO_MAVEN="${{ inputs.upload_to_maven }}"
         if [[ "$UPLOAD_TO_MAVEN" == "true" ]]; then
-          (cd aar-out; ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:publishToMavenCentral)
+          (cd extension/android; ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:publishToMavenCentral)
         fi
 
   upload-release-aar:

CMakeLists.txt

@@ -761,12 +761,16 @@ if(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/flat_tensor)
 endif()
 
+if(EXECUTORCH_BUILD_EXTENSION_MODULE)
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/module)
+endif()
+
 if(EXECUTORCH_BUILD_EXTENSION_LLM)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/llm/tokenizers)
 endif()
 
-if(EXECUTORCH_BUILD_EXTENSION_MODULE)
-  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/module)
+if(EXECUTORCH_BUILD_EXTENSION_LLM_RUNNER)
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/llm/runner)
 endif()
 
 if(EXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL)
@@ -810,6 +814,10 @@ if(EXECUTORCH_BUILD_PYBIND)
       torch
   )
 
+  if(EXECUTORCH_BUILD_TESTS)
+    list(APPEND _dep_libs test_backend_compiler_lib)
+  endif()
+
   if(EXECUTORCH_BUILD_KERNELS_OPTIMIZED)
     list(APPEND _dep_libs optimized_native_cpu_ops_lib)
   else()

CONTRIBUTING.md

@@ -5,10 +5,10 @@ it easy to contribute to this project.
 ## Dev Install
 
 Set up your environment by following the instructions at
-https://pytorch.org/executorch/stable/getting-started-setup.html to clone
+https://pytorch.org/executorch/main/getting-started-setup to clone
 the repo and install the necessary requirements.
 
-Refer to this [document](https://pytorch.org/executorch/main/using-executorch-building-from-source.html) to build ExecuTorch from source.
+Refer to this [document](docs/source/using-executorch-building-from-source.md) to build ExecuTorch from source.
 
 ### Dev Setup for Android
 For Android, please refer to the [Android documentation](docs/source/using-executorch-android.md).
@@ -40,12 +40,12 @@ executorch
 ├── <a href="devtools">devtools</a> - Model profiling, debugging, and inspection. Please refer to the <a href="docs/source/devtools-overview.md">tools documentation</a> for more information.
 │   ├── <a href="devtools/bundled_program">bundled_program</a> - a tool for validating ExecuTorch model. See <a href="docs/source/bundled-io.md">doc</a>.
 │   ├── <a href="devtools/etdump">etdump</a> - ETDump - a format for saving profiling and debugging data from runtime. See <a href="docs/source/etdump.md">doc</a>.
-│   ├── <a href="devtools/etrecord">etrecord</a> - ETRecord - AOT debug artifact for ExecuTorch. See <a href="https://pytorch.org/executorch/main/etrecord.html">doc</a>.
-│   ├── <a href="devtools/inspector">inspector</a> - Python API to inspect ETDump and ETRecord. See <a href="https://pytorch.org/executorch/main/model-inspector.html">doc</a>.
+│   ├── <a href="devtools/etrecord">etrecord</a> - ETRecord - AOT debug artifact for ExecuTorch. See <a href="https://pytorch.org/executorch/main/etrecord">doc</a>.
+│   ├── <a href="devtools/inspector">inspector</a> - Python API to inspect ETDump and ETRecord. See <a href="https://pytorch.org/executorch/main/model-inspector">doc</a>.
 │   └── <a href="devtools/visualization">visualization</a> - Visualization tools for representing model structure and performance metrics.
 ├── <a href="docs">docs</a> - Static docs tooling and documentation source files.
 ├── <a href="examples">examples</a> - Examples of various user flows, such as model export, delegates, and runtime execution.
-├── <a href="exir">exir</a> - Ahead-of-time library: model capture and lowering APIs. EXport Intermediate Representation (EXIR) is a format for representing the result of <a href="https://pytorch.org/docs/main/export.ir_spec.html">torch.export</a>. This directory contains utilities and passes for lowering the EXIR graphs into different <a href="/docs/source/ir-exir.md">dialects</a> and eventually suitable to run on target hardware.
+├── <a href="exir">exir</a> - Ahead-of-time library: model capture and lowering APIs. EXport Intermediate Representation (EXIR) is a format for representing the result of <a href="https://pytorch.org/docs/stable/export.html">torch.export</a>. This directory contains utilities and passes for lowering the EXIR graphs into different <a href="/docs/source/ir-exir.md">dialects</a> and eventually suitable to run on target hardware.
 │   ├── <a href="exir/_serialize">_serialize</a> - Serialize final export artifact.
 │   ├── <a href="exir/backend">backend</a> - Backend delegate ahead of time APIs.
 │   ├── <a href="exir/capture">capture</a> - Program capture.
@@ -57,8 +57,8 @@ executorch
 │   ├── <a href="exir/serde">serde</a> - Graph module serialization/deserialization.
 │   ├── <a href="exir/verification">verification</a> - IR verification.
 ├── <a href="extension">extension</a> - Extensions built on top of the runtime.
-│   ├── <a href="extension/android">android</a> - ExecuTorch wrappers for Android apps. Please refer to the <a href="docs/source/using-executorch-android.md">Android documentation</a> and <a href="https://pytorch.org/executorch/main/javadoc/">Javadoc</a> for more information.
-│   ├── <a href="extension/apple">apple</a> - ExecuTorch wrappers for iOS apps. Please refer to the <a href="docs/source/using-executorch-ios.md">iOS documentation</a> and <a href="https://pytorch.org/executorch/stable/apple-runtime.html">how to integrate into Apple platform</a> for more information.
+│   ├── <a href="extension/android">android</a> - ExecuTorch wrappers for Android apps. Please refer to the <a href="docs/source/using-executorch-android.md">Android documentation</a> and <a href="https://pytorch.org/executorch/main/javadoc">Javadoc</a> for more information.
+│   ├── <a href="extension/apple">apple</a> - ExecuTorch wrappers for iOS apps. Please refer to the <a href="docs/source/using-executorch-ios.md">iOS documentation</a> on how to integrate into Apple platform</a> for more information.
 │   ├── <a href="extension/aten_util">aten_util</a> - Converts to and from PyTorch ATen types.
 │   ├── <a href="extension/data_loader">data_loader</a> - 1st party data loader implementations.
 │   ├── <a href="extension/evalue_util">evalue_util</a> - Helpers for working with EValue objects.
@@ -68,10 +68,10 @@ executorch
 │   ├── <a href="extension/memory_allocator">memory_allocator</a> - 1st party memory allocator implementations.
 │   ├── <a href="extension/module">module</a> - A simplified C++ wrapper for the runtime. An abstraction that deserializes and executes an ExecuTorch artifact (.pte file). Refer to the <a href="docs/source/extension-module.md">module documentation</a> for more information.
 │   ├── <a href="extension/parallel">parallel</a> - C++ threadpool integration.
-│   ├── <a href="extension/pybindings">pybindings</a> - Python API for executorch runtime. This is powering up the <a href="https://pytorch.org/executorch/main/runtime-python-api-reference.html">runtime Python API</a> for ExecuTorch.
+│   ├── <a href="extension/pybindings">pybindings</a> - Python API for executorch runtime. This is powering up the <a href="docs/source/runtime-python-api-reference.md">runtime Python API</a> for ExecuTorch.
 │   ├── <a href="extension/pytree">pytree</a> - C++ and Python flattening and unflattening lib for pytrees.
 │   ├── <a href="extension/runner_util">runner_util</a> - Helpers for writing C++ PTE-execution tools.
-│   ├── <a href="extension/tensor">tensor</a> - Tensor maker and <code>TensorPtr</code>, details in <a href="/docs/source/extension-tensor.md">this documentation</a>. For how to use <code>TensorPtr</code> and <code>Module</code>, please refer to the <a href="/docs/source/using-executorch-cpp.md">"Using ExecuTorch with C++"</a> doc.
+│   ├── <a href="extension/tensor">tensor</a> - Tensor maker and <code>TensorPtr</code>, details in <a href="docs/source/extension-tensor.md">this documentation</a>. For how to use <code>TensorPtr</code> and <code>Module</code>, please refer to the <a href="docs/source/using-executorch-cpp.md">"Using ExecuTorch with C++"</a> doc.
 │   ├── <a href="extension/testing_util">testing_util</a> - Helpers for writing C++ tests.
 │   ├── <a href="extension/threadpool">threadpool</a> - Threadpool.
 │   └── <a href="extension/training">training</a> - Experimental libraries for on-device training.
@@ -85,7 +85,7 @@ executorch
 ├── <a href="runtime">runtime</a> - Core C++ runtime. These components are used to execute the ExecuTorch program. Please refer to the <a href="docs/source/runtime-overview.md">runtime documentation</a> for more information.
 │   ├── <a href="runtime/backend">backend</a> - Backend delegate runtime APIs.
 │   ├── <a href="runtime/core">core</a> - Core structures used across all levels of the runtime. Basic components such as <code>Tensor</code>, <code>EValue</code>, <code>Error</code> and <code>Result</code> etc.
-│   ├── <a href="runtime/executor">executor</a> - Model loading, initialization, and execution. Runtime components that execute the ExecuTorch program, such as <code>Program</code>, <code>Method</code>. Refer to the <a href="https://pytorch.org/executorch/main/executorch-runtime-api-reference.html">runtime API documentation</a> for more information.
+│   ├── <a href="runtime/executor">executor</a> - Model loading, initialization, and execution. Runtime components that execute the ExecuTorch program, such as <code>Program</code>, <code>Method</code>. Refer to the <a href="https://pytorch.org/executorch/main/executorch-runtime-api-reference">runtime API documentation</a> for more information.
 │   ├── <a href="runtime/kernel">kernel</a> - Kernel registration and management.
 │   └── <a href="runtime/platform">platform</a> - Layer between architecture specific code and portable C++.
 ├── <a href="schema">schema</a> - ExecuTorch PTE file format flatbuffer schemas.
@@ -102,7 +102,7 @@ executorch
 ## Contributing workflow
 We actively welcome your pull requests (PRs).
 
-If you're completely new to open-source projects, GitHub, or ExecuTorch, please see our [New Contributor Guide](./docs/source/new-contributor-guide.md) for a step-by-step walkthrough on making your first contribution. Otherwise, read on.
+If you're completely new to open-source projects, GitHub, or ExecuTorch, please see our [New Contributor Guide](docs/source/new-contributor-guide.md) for a step-by-step walkthrough on making your first contribution. Otherwise, read on.
 
 1. [Claim an issue](#claiming-issues), if present, before starting work. If an
    issue doesn't cover the work you plan to do, consider creating one to provide
@@ -245,7 +245,7 @@ modifications to the Google C++ style guide.
 
 ### C++ Portability Guidelines
 
-See also [Portable C++ Programming](/docs/source/portable-cpp-programming.md)
+See also [Portable C++ Programming](docs/source/portable-cpp-programming.md)
 for detailed advice.
 
 #### C++ language version
@@ -417,9 +417,9 @@ for basics.
 
 ## For Backend Delegate Authors
 
-- Use [this](/docs/source/backend-delegates-integration.md) guide when
+- Use [this](docs/source/backend-delegates-integration.md) guide when
   integrating your delegate with ExecuTorch.
-- Refer to [this](/docs/source/backend-delegates-dependencies.md) set of
+- Refer to [this](docs/source/backend-delegates-dependencies.md) set of
   guidelines when including a third-party depenency for your delegate.
 
 &nbsp;

Package.swift

@@ -15,7 +15,7 @@
 //
 // For details on building frameworks locally or using prebuilt binaries,
 // see the documentation:
-// https://pytorch.org/executorch/main/using-executorch-ios.html
+// https://pytorch.org/executorch/main/using-executorch-ios
 
 import PackageDescription