Merge branch 'pytorch:main' into pr_model_improve

Author: Michiel-Olieslagers

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-7ae0ce6360b6e4f944906502d20da24c04debee5
+59d5cf083b4f860dea76fe8936076177f9367f10

.ci/scripts/setup-openvino.sh

@@ -0,0 +1,28 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+set -ex
+
+# shellcheck source=/dev/null
+source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
+
+git clone https://github.com/openvinotoolkit/openvino.git
+cd openvino && git checkout releases/2025/1
+git submodule update --init --recursive
+sudo ./install_build_dependencies.sh
+mkdir build && cd build
+cmake .. -DCMAKE_BUILD_TYPE=Release -DENABLE_PYTHON=ON
+make -j$(nproc)
+
+cd ..
+cmake --install build --prefix dist
+
+source dist/setupvars.sh
+cd ../backends/openvino
+pip install -r requirements.txt
+cd scripts
+./openvino_build.sh --enable_python

.ci/scripts/test_llama.sh

@@ -269,7 +269,7 @@ $PYTHON_EXECUTABLE -m examples.models.llama.export_llama ${EXPORT_ARGS}
 
 # Create tokenizer.bin.
 echo "Creating tokenizer.bin"
-$PYTHON_EXECUTABLE -m extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
+$PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
 
 
 RUNTIME_ARGS="--model_path=${EXPORTED_MODEL_NAME} --tokenizer_path=tokenizer.bin --prompt=Once --temperature=0 --seq_len=10 --warmup=1"

.ci/scripts/test_llama_torchao_lowbit.sh

@@ -55,7 +55,7 @@ cmake --build cmake-out/examples/models/llama -j16 --config Release
 download_stories_model_artifacts
 
 echo "Creating tokenizer.bin"
-$PYTHON_EXECUTABLE -m extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
+$PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
 
 # Export model
 LLAMA_CHECKPOINT=stories110M.pt

.ci/scripts/test_openvino.sh

@@ -0,0 +1,16 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+set -ex
+
+# shellcheck source=/dev/null
+source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
+
+source openvino/dist/setupvars.sh
+cd backends/openvino/tests
+python test_runner.py --test_type ops
+python test_runner.py --test_type models

.ci/scripts/test_phi_3_mini.sh

@@ -56,7 +56,7 @@ cmake_build_phi_3_mini() {
 prepare_tokenizer() {
   echo "Downloading and converting tokenizer.model"
   wget -O tokenizer.model "https://huggingface.co/microsoft/Phi-3-mini-128k-instruct/resolve/main/tokenizer.model?download=true"
-  $PYTHON_EXECUTABLE -m executorch.extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
+  $PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
 }
 
 # Export phi-3-mini model to pte

.ci/scripts/test_qnn_static_llama.sh

@@ -5,7 +5,7 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
-set -exu
+set -euxo pipefail
 
 source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
 
@@ -30,7 +30,7 @@ pip install graphviz
 # Download stories llama110m artifacts
 download_stories_model_artifacts
 echo "Creating tokenizer.bin"
-$PYTHON_EXECUTABLE -m extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
+$PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
 
 set +e
 # Compile only as weight sharing is not applicable on x86
@@ -56,4 +56,3 @@ if [ $exit_code1 -ne 0 ] || [ $exit_code2 -ne 0 ]; then
 else
     exit 0
 fi
-set -e

.ci/scripts/utils.sh

@@ -32,7 +32,7 @@ install_executorch() {
   which pip
   # Install executorch, this assumes that Executorch is checked out in the
   # current directory.
-  ./install_executorch.sh --pybind xnnpack "$@"
+  ./install_executorch.sh "$@"
   # Just print out the list of packages for debugging
   pip list
 }

.github/release.yml

@@ -0,0 +1,96 @@
+# .github/release.yml
+
+changelog:
+  exclude:
+    labels:
+      - ignore-for-release
+  categories:
+    - title: Breaking Changes 🛠
+      labels:
+        - Semver-Major
+        - breaking-change
+    - title: API
+      labels:
+        - "release notes: api"
+    - title: ARM
+      labels:
+        - "release notes: arm"
+        - "module: arm"
+        - "partner: arm"
+    - title: NXP
+      labels:
+        - "release notes: nxp"
+        - "module: nxp"
+    - title: Exir
+      labels:
+        - "release notes: exir"
+        - "module: exir"
+    - title: Misc
+      labels:
+        - "release notes: misc"
+    - title: Apple
+      labels:
+        - "release notes: apple"
+        - "module: coreml"
+        - "module: mps"
+    - title: Android
+      labels:
+        - "module: android"
+    - title: IOS
+      labels:
+        - "module: ios"
+    - title: Build
+      labels:
+        - "release notes: build"
+    - title: Vulkan
+      labels:
+        - "release notes: vulkan"
+        - "module: vulkan"
+    - title: Cadence
+      labels:
+        - "release notes: cadence"
+        - "module: cadence"
+    - title: Runtime
+      labels:
+        - "release notes: runtime"
+        - "module: runtime"
+    - title: XNNPACK
+      labels:
+        - "release notes: xnnpack"
+        - "module: xnnpack"
+    - title: Devtools
+      labels:
+        - "release notes: devtools"   
+        - "module: devtools"
+    - title: Examples
+      labels:
+        - "release notes: examples"
+    - title: LLM
+      labels:
+        - "module: llm"
+    - title: Mediatek
+      labels:
+        - "release notes: mediatek"
+        - "partner: mediatek"
+    - title: Openvino
+      labels:
+        - "release notes: openvino"
+    - title: Qualcomm
+      labels:
+        - "release notes: qualcomm"
+        - "partner: qualcomm"
+        - "module: qnn"
+    - title: Training
+      labels:
+        - "release notes: training"
+        - "module: training"
+    - title: Quantization
+      labels:
+        - "release notes: quantization" 
+    - title: Ops & kernels
+      labels:
+        - "release notes: ops & kernels"
+        - "module: kernels"
+    - title: Other Changes
+      labels:
+        - "*"

.github/scripts/extract_benchmark_results.py

@@ -360,6 +360,7 @@ def transform(
                     "app_type": app_type,
                     # Just keep a copy of the benchmark config here
                     "benchmark_config": json.dumps(benchmark_config),
+                    "job_conclusion": "SUCCESS",
                 },
             },
             "model": {
@@ -455,7 +456,7 @@ def transform_failure_record(
         },
         "metric": {
             "name": "FAILURE_REPORT",
-            "benchmark_values": 0,
+            "benchmark_values": [0],
             "target_value": 0,
             "extra_info": {
                 "method": "",