Update

[ghstack-poisoned]

Author: swolchok

.ci/scripts/test_eval_llama_mmlu.sh

@@ -43,6 +43,7 @@ run_and_verify() {
 	--tasks mmlu \
 	-f 5 \
 	--max_seq_length 2048 \
+	--max_context_length 2048 \
 	--limit 5 > result.txt
 
     # Verify result.txt

.ci/scripts/test_eval_llama_wikitext.sh

@@ -41,6 +41,7 @@ run_and_verify() {
 	-kv \
 	-d fp32 \
 	--max_seq_length 2048 \
+	--max_context_length 2048 \
 	--limit 5 > result.txt
 
     # Verify result.txt

.github/workflows/apple.yml

@@ -37,7 +37,7 @@ jobs:
         id: set_version
         shell: bash
         run: |
-          VERSION="0.4.0.$(TZ='PST8PDT' date +%Y%m%d)"
+          VERSION="0.5.0.$(TZ='PST8PDT' date +%Y%m%d)"
           echo "version=$VERSION" >> "$GITHUB_OUTPUT"
 
   build-demo-ios:

.github/workflows/pull.yml

@@ -221,7 +221,7 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "cmake"
 
         # install pybind
-        bash install_executorch.sh --pybind xnnpack
+        bash install_executorch.sh --pybind xnnpack --use-pt-pinned-commit
 
         # install Llava requirements
         bash examples/models/llama/install_requirements.sh
@@ -484,7 +484,7 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "cmake"
 
         # install pybind
-        bash install_executorch.sh --pybind xnnpack
+        bash install_executorch.sh --pybind xnnpack --use-pt-pinned-commit
 
         # install phi-3-mini requirements
         bash examples/models/phi-3-mini/install_requirements.sh
@@ -514,7 +514,7 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "cmake"
 
         # install pybind
-        bash install_executorch.sh --pybind xnnpack
+        bash install_executorch.sh --pybind xnnpack --use-pt-pinned-commit
 
         # install llama requirements
         bash examples/models/llama/install_requirements.sh
@@ -544,7 +544,7 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "cmake"
 
         # install pybind
-        bash install_executorch.sh --pybind xnnpack
+        bash install_executorch.sh --pybind xnnpack --use-pt-pinned-commit
 
         # install llama requirements
         bash examples/models/llama/install_requirements.sh
@@ -574,7 +574,7 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "cmake"
 
         # install pybind
-        bash install_executorch.sh --pybind xnnpack
+        bash install_executorch.sh --pybind xnnpack --use-pt-pinned-commit
 
         # install llama requirements
         bash examples/models/llama/install_requirements.sh

.github/workflows/trunk.yml

@@ -150,7 +150,7 @@ jobs:
         conda activate "${CONDA_ENV}"
 
         source .ci/scripts/utils.sh
-        install_executorch
+        install_executorch "use-pt-pinned-commit"
 
         .ci/scripts/setup-arm-baremetal-tools.sh
 
@@ -180,7 +180,7 @@ jobs:
         conda activate "${CONDA_ENV}"
 
         source .ci/scripts/utils.sh
-        install_executorch
+        install_executorch "use-pt-pinned-commit"
 
         .ci/scripts/setup-arm-baremetal-tools.sh
 

.lintrunner.toml

@@ -1,4 +1,4 @@
-merge_base_with = "origin/main"
+merge_base_with = "main"
 
 [[linter]]
 code = 'FLAKE8'

CMakeLists.txt

@@ -1,4 +1,5 @@
 # Copyright (c) Meta Platforms, Inc. and affiliates.
+# Copyright 2024-2025 Arm Limited and/or its affiliates.
 # All rights reserved.
 #
 # This source code is licensed under the BSD-style license found in the
@@ -819,6 +820,14 @@ if(EXECUTORCH_BUILD_EXECUTOR_RUNNER)
     list(APPEND _executor_runner_libs quantized_ops_lib)
   endif()
 
+  if(EXECUTORCH_ENABLE_EVENT_TRACER)
+    if(EXECUTORCH_BUILD_DEVTOOLS)
+      list(APPEND _executor_runner_libs etdump flatccrt)
+    else()
+      message(SEND_ERROR "Use of 'EXECUTORCH_ENABLE_EVENT_TRACER' requires 'EXECUTORCH_BUILD_DEVTOOLS' to be enabled.")
+    endif()
+  endif()
+
   add_executable(executor_runner ${_executor_runner__srcs})
   if(CMAKE_BUILD_TYPE STREQUAL "Release")
     if(APPLE)

CONTRIBUTING.md

@@ -44,6 +44,38 @@ Meta has a [bounty program](https://www.facebook.com/whitehat/) for the safe
 disclosure of security bugs. In those cases, please go through the process
 outlined on that page and do not file a public issue.
 
+### Issue Labels
+
+#### Module/Partner Labels
+
+[Labels beginning with `module:`](https://github.com/pytorch/executorch/labels?q=%22module%3A+%22)
+indicate the area that the issue relates to. The ExecuTorch oncall will
+typically add this label.
+
+[Labels beginning with `partner:`](https://github.com/pytorch/executorch/labels?q=%22partner%3A+%22)
+indicate the ExecuTorch partner who owns the issue. The ExecuTorch oncall will
+typically add this label.
+
+#### Lifecycle Labels
+
+The ExecuTorch oncall will triage new issues. If the issue requires more
+information from the issue's author, oncall will add the `need-user-input` label
+and wait for the author to respond.
+
+Once the issue contains enough information, the oncall will:
+- Ensure that the title is descriptive
+- Add one of the labels:
+  - `bug`: The issue describes an unexpected problem
+  - `feature`: The issue describes a request for new functionality
+  - `rfc`: The issue describes a proposed change to functionality
+- Add one `module:` label or one `partner:` label, as described above
+- Add the `triaged` label
+
+After this point, the oncall has finished the triage process, and the
+module owner or partner is responsible for resolving the issue. (See
+https://github.com/pytorch/executorch/issues/7679 for the mapping of labels to
+owners.)
+
 ### Claiming Issues
 We'd love your help closing out [open
 issues](https://github.com/pytorch/executorch/issues?q=sort%3Aupdated-desc+is%3Aissue+is%3Aopen)

README-wheel.md

@@ -4,20 +4,21 @@ standard on-device iOS and Android mobile deployments. One of the main goals for
 ExecuTorch is to enable wider customization and deployment capabilities of the
 PyTorch programs.
 
-The `executorch` pip package is in alpha.
-* Supported python versions: 3.10, 3.11
+The `executorch` pip package is in beta.
+* Supported python versions: 3.10, 3.11, 3.12
 * Compatible systems: Linux x86_64, macOS aarch64
 
-The prebuilt `executorch.extension.pybindings.portable_lib` module included in
-this package provides a way to run ExecuTorch `.pte` files, with some
-restrictions:
+The prebuilt `executorch.runtime` module included in this package provides a way
+to run ExecuTorch `.pte` files, with some restrictions:
 * Only [core ATen
   operators](https://pytorch.org/executorch/stable/ir-ops-set-definition.html)
   are linked into the prebuilt module
 * Only the [XNNPACK backend
   delegate](https://pytorch.org/executorch/main/native-delegates-executorch-xnnpack-delegate.html)
-  is linked into the prebuilt module
-* [macOS only] [Core ML](https://pytorch.org/executorch/main/build-run-coreml.html) and [MPS](https://pytorch.org/executorch/main/build-run-mps.html) backend delegates are linked into the prebuilt module.
+  is linked into the prebuilt module.
+* \[macOS only] [Core ML](https://pytorch.org/executorch/main/build-run-coreml.html)
+  and [MPS](https://pytorch.org/executorch/main/build-run-mps.html) backend
+  delegates are also linked into the prebuilt module.
 
 Please visit the [ExecuTorch website](https://pytorch.org/executorch/) for
 tutorials and documentation. Here are some starting points:

backends/arm/README.md

@@ -122,6 +122,18 @@ The you can run the tests with
 pytest -c /dev/null -v -n auto backends/arm/test --arm_run_corstoneFVP
 ```
 
+## Passes
+
+With the default passes in the Arm Ethos-U backend, assuming the model lowers fully to the
+Ethos-U, the exported program is composed of a Quantize node, Ethos-U custom delegate
+and a Dequantize node. In some circumstances, you may want to feed quantized input to the Neural
+Network straight away, e.g. if you have a camera sensor outputting (u)int8 data and keep all the
+arithmetic of the application in the int8 domain. For these cases, you can apply the
+`exir/passes/quantize_io_pass.py`. See the unit test in `executorch/backends/arm/
+test/passes/test_ioquantization_pass.py`for an example how to feed quantized inputs and
+obtain quantized outputs.
+
+
 ### Code coverage
 
 To get code coverage: