Skip to content

Commit a194264

Browse files
nil-is-allnil-is-all
authored
Merge branch 'main' into nil-is-all-patch-1
2 parents 7fff2bf + 8a0a25b commit a194264

File tree

938 files changed

+43130
-11049
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

938 files changed

+43130
-11049
lines changed

‎.ci/docker/ci_commit_pins/pytorch.txt‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1 +1 @@
1-
e7152ff8a6a929a0db7f3f4a72a5b6d471769cd3
1+
4d4abec80f03cd8fdefe1d9cb3a60d3690cd777e

‎.ci/scripts/setup-qnn-deps.sh‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -11,4 +11,4 @@ source "$(dirname "${BASH_SOURCE[0]}")/../../backends/qualcomm/scripts/install_q
1111

1212
setup_libcpp 12
1313
setup_android_ndk
14-
install_qnn
14+
install_qnn

‎.ci/scripts/setup-samsung-linux-deps.sh‎

Lines changed: 68 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,68 @@
1+
#!/bin/bash
2+
# Copyright (c) Meta Platforms, Inc. and affiliates.
3+
# Copyright (c) Samsung Electronics Co. LTD
4+
# All rights reserved.
5+
#
6+
# This source code is licensed under the BSD-style license found in the
7+
# LICENSE file in the root directory of this source tree.
8+
9+
set -ex
10+
11+
12+
download_ai_lite_core() {
13+
API_BASE="https://soc-developer.semiconductor.samsung.com/api/v1/resource/ai-litecore/download"
14+
API_KEY="kn10SoSY3hkC-9Qny5TqD2mnqVrlupv3krnjLeBt5cY"
15+
16+
VERSION="0.5"
17+
OS_NAME="Ubuntu 22.04"
18+
OUT_FILE="/tmp/exynos-ai-litecore-v${VERSION}.tar.gz"
19+
TARGET_PATH="/tmp/exynos_ai_lite_core"
20+
21+
mkdir -p ${TARGET_PATH}
22+
# Presigned issue URL
23+
JSON_RESP=$(curl -sS -G \
24+
--location --fail --retry 3 \
25+
-H "apikey: ${API_KEY}" \
26+
--data-urlencode "version=${VERSION}" \
27+
--data-urlencode "os=${OS_NAME}" \
28+
"${API_BASE}")
29+
30+
DOWNLOAD_URL=$(echo "$JSON_RESP" | sed -n 's/.*"data":[[:space:]]*"\([^"]*\)".*/\1/p')
31+
32+
if [[ -z "$DOWNLOAD_URL" ]]; then
33+
echo "Failed to extract download URL"
34+
echo "$JSON_RESP"
35+
exit 1
36+
fi
37+
38+
# Download LiteCore
39+
curl -sS -L --fail --retry 3 \
40+
--output "$OUT_FILE" \
41+
"$DOWNLOAD_URL"
42+
43+
echo "Download done: $OUT_FILE"
44+
45+
46+
tar -C "${TARGET_PATH}" --strip-components=1 -xzvf "${OUT_FILE}"
47+
48+
export EXYNOS_AI_LITECORE_ROOT=${TARGET_PATH}
49+
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH:-}:${EXYNOS_AI_LITECORE_ROOT}/lib/x86_64-linux
50+
}
51+
52+
install_enn_backend() {
53+
NDK_INSTALLATION_DIR=/opt/ndk
54+
rm -rf "${NDK_INSTALLATION_DIR}" && sudo mkdir -p "${NDK_INSTALLATION_DIR}"
55+
ANDROID_NDK_VERSION=r27b
56+
57+
# build Exynos backend
58+
export ANDROID_NDK_ROOT=${ANDROID_NDK_ROOT:-/opt/ndk}
59+
bash backends/samsung/build.sh --build all
60+
# set env variable
61+
export EXECUTORCH_ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd)"
62+
export PYTHONPATH=${PYTHONPATH:-}:${EXECUTORCH_ROOT}/..
63+
}
64+
65+
AI_LITE_CORE_VERSION=0.5.0
66+
67+
download_ai_lite_core ${AI_LITE_CORE_VERSION}
68+
install_enn_backend

‎.ci/scripts/setup-windows.ps1‎

Lines changed: 24 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,24 @@
1+
param (
2+
[string]$editable = $false
3+
)
4+
5+
conda create --yes --quiet -n et python=3.12
6+
conda activate et
7+
8+
# Activate the VS environment - this is required for Dynamo to work, as it uses MSVC.
9+
# There are a bunch of environment variables that it requires.
10+
# See https://learn.microsoft.com/en-us/cpp/build/building-on-the-command-line.
11+
& "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\Common7\Tools\Launch-VsDevShell.ps1" -Arch amd64
12+
13+
# Install test dependencies
14+
pip install -r .ci/docker/requirements-ci.txt
15+
16+
if ($editable -eq 'true') {
17+
install_executorch.bat --editable
18+
} else {
19+
install_executorch.bat
20+
}
21+
if ($LASTEXITCODE -ne 0) {
22+
Write-Host "Installation was unsuccessful. Exit code: $LASTEXITCODE."
23+
exit $LASTEXITCODE
24+
}

‎.ci/scripts/test_backend_linux.sh‎

Lines changed: 1 addition & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -30,10 +30,8 @@ if [[ "$FLOW" == *qnn* ]]; then
3030
# Qualcomm build. TODO (gjcomer) Clean this up once the QNN pybinding integration is
3131
# cleaned up.
3232
PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh --build-tool cmake
33-
PYTHON_EXECUTABLE=python bash .ci/scripts/setup-qnn-deps.sh
34-
PYTHON_EXECUTABLE=python bash .ci/scripts/build-qnn-sdk.sh
33+
PYTHON_EXECUTABLE=python source .ci/scripts/build-qnn-sdk.sh
3534
QNN_X86_LIB_DIR=`realpath build-x86/lib/`
36-
QNN_SDK_ROOT="/tmp/qnn/2.28.0.241029"
3735
export LD_LIBRARY_PATH"=$QNN_X86_LIB_DIR:$QNN_SDK_ROOT/lib/x86_64-linux-clang/:${LD_LIBRARY_PATH:-}"
3836

3937
# TODO Get SDK root from install scripts

‎.ci/scripts/test_model.ps1‎

Lines changed: 89 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,89 @@
1+
param (
2+
[string]$modelName,
3+
[string]$backend,
4+
[string]$buildDir = "cmake-out",
5+
[bool]$strict = $false
6+
)
7+
8+
Set-PSDebug -Trace 1
9+
$ErrorActionPreference = 'Stop'
10+
$PSNativeCommandUseErrorActionPreference = $true
11+
12+
function ExportModel-Portable {
13+
param (
14+
[string]$model_name,
15+
[bool]$strict
16+
)
17+
18+
$exportParams = "--model_name", "$modelName"
19+
if ($strict) {
20+
$exportParams += "--strict"
21+
}
22+
python -m examples.portable.scripts.export @exportParams | Write-Host
23+
if ($LASTEXITCODE -ne 0) {
24+
Write-Host "Model export failed. Exit code: $LASTEXITCODE."
25+
exit $LASTEXITCODE
26+
}
27+
28+
"$modelName.pte"
29+
}
30+
31+
function ExportModel-Xnnpack {
32+
param (
33+
[string]$model_name,
34+
[bool]$quantize
35+
)
36+
37+
if ($quantize) {
38+
python -m examples.xnnpack.aot_compiler --model_name="${MODEL_NAME}" --delegate --quantize | Write-Host
39+
$modelFile = "$($modelName)_xnnpack_q8.pte"
40+
} else {
41+
python -m examples.xnnpack.aot_compiler --model_name="${MODEL_NAME}" --delegate | Write-Host
42+
$modelFile = "$($modelName)_xnnpack_fp32.pte"
43+
}
44+
if ($LASTEXITCODE -ne 0) {
45+
Write-Host "Model export failed. Exit code: $LASTEXITCODE."
46+
exit $LASTEXITCODE
47+
}
48+
49+
$modelFile
50+
}
51+
52+
# Build the runner
53+
if (Test-Path -Path $buildDir) {
54+
Remove-Item -Path $buildDir -Recurse -Force
55+
}
56+
New-Item -Path $buildDir -ItemType Directory
57+
Push-Location $buildDir
58+
cmake .. --preset windows
59+
cmake --build . -t executor_runner -j16 --config Release
60+
if ($LASTEXITCODE -ne 0) {
61+
Write-Host "Runner build failed. Exit code: $LASTEXITCODE."
62+
exit $LASTEXITCODE
63+
}
64+
$executorBinaryPath = Join-Path -Path $buildDir -ChildPath "Release\executor_runner.exe"
65+
Pop-Location
66+
67+
# Export the model
68+
switch ($backend) {
69+
"portable" {
70+
$model_path = ExportModel-Portable -model_name $modelName -strict $strict
71+
}
72+
"xnnpack-f32" {
73+
$model_path = ExportModel-Xnnpack -model_name $modelName -quantize $false
74+
}
75+
"xnnpack-q8" {
76+
$model_path = ExportModel-Xnnpack -model_name $modelName -quantize $true
77+
}
78+
default {
79+
Write-Host "Unknown backend $backend."
80+
exit 1
81+
}
82+
}
83+
84+
# Run the runner
85+
& "$executorBinaryPath" --model_path="$model_path"
86+
if ($LASTEXITCODE -ne 0) {
87+
Write-Host "Model execution failed. Exit code: $LASTEXITCODE."
88+
exit $LASTEXITCODE
89+
}

‎.ci/scripts/test_model.sh‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -97,7 +97,7 @@ test_model() {
9797
bash examples/models/llava/install_requirements.sh
9898
STRICT="--no-strict"
9999
fi
100-
if [[ "${MODEL_NAME}" == "qwen2_5" ]]; then
100+
if [[ "${MODEL_NAME}" == "qwen2_5_1_5b" ]]; then
101101
# Install requirements for export_llama
102102
bash examples/models/llama/install_requirements.sh
103103
# Test export_llm script: python3 -m extension.llm.export.export_llm.

‎.ci/scripts/test_qnn_static_llama.sh‎

Lines changed: 0 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -9,8 +9,6 @@ set -euxo pipefail
99

1010
source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
1111

12-
# Source QNN configuration
13-
source "$(dirname "${BASH_SOURCE[0]}")/../../backends/qualcomm/scripts/qnn_config.sh"
1412
# Download QNN_SDK. If already downloaded, export environment path
1513
source "$(dirname "${BASH_SOURCE[0]}")/../../backends/qualcomm/scripts/install_qnn_sdk.sh"
1614
install_qnn

‎.ci/scripts/test_torchao_huggingface_checkpoints.sh‎

Lines changed: 139 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,139 @@
1+
#!/usr/bin/env bash
2+
set -euo pipefail
3+
4+
# -------------------------
5+
# Args / flags
6+
# -------------------------
7+
TEST_WITH_RUNNER=0
8+
MODEL_NAME=""
9+
10+
# Parse args
11+
if [[ $# -lt 1 ]]; then
12+
echo "Usage: $0 <model_name> [--test_with_runner]"
13+
echo "Supported model_name values: qwen3_4b, phi_4_mini"
14+
exit 1
15+
fi
16+
17+
MODEL_NAME="$1"
18+
shift
19+
20+
while [[ $# -gt 0 ]]; do
21+
case "$1" in
22+
--test_with_runner)
23+
TEST_WITH_RUNNER=1
24+
;;
25+
-h|--help)
26+
echo "Usage: $0 <model_name> [--test_with_runner]"
27+
echo " model_name: qwen3_4b | phi_4_mini"
28+
echo " --test_with_runner: build ET + run llama_main to sanity-check the export"
29+
exit 0
30+
;;
31+
*)
32+
echo "Unknown option: $1"
33+
exit 1
34+
;;
35+
esac
36+
shift
37+
done
38+
39+
if [[ -z "${PYTHON_EXECUTABLE:-}" ]]; then
40+
PYTHON_EXECUTABLE=python3
41+
fi
42+
43+
MODEL_OUT=model.pte
44+
45+
case "$MODEL_NAME" in
46+
qwen3_4b)
47+
echo "Running Qwen3-4B export..."
48+
HF_MODEL_DIR=$(hf download pytorch/Qwen3-4B-INT8-INT4)
49+
EXPECTED_MODEL_SIZE_UPPER_BOUND=$((3 * 1024 * 1024 * 1024)) # 3GB
50+
$PYTHON_EXECUTABLE -m executorch.examples.models.qwen3.convert_weights \
51+
$HF_MODEL_DIR \
52+
pytorch_model_converted.bin
53+
54+
$PYTHON_EXECUTABLE -m executorch.examples.models.llama.export_llama \
55+
--model "qwen3_4b" \
56+
--checkpoint pytorch_model_converted.bin \
57+
--params examples/models/qwen3/config/4b_config.json \
58+
--output_name $MODEL_OUT \
59+
-kv \
60+
--use_sdpa_with_kv_cache \
61+
-X \
62+
--xnnpack-extended-ops \
63+
--max_context_length 1024 \
64+
--max_seq_length 1024 \
65+
--dtype fp32 \
66+
--metadata '{"get_bos_id":199999, "get_eos_ids":[200020,199999]}'
67+
;;
68+
69+
phi_4_mini)
70+
echo "Running Phi-4-mini export..."
71+
HF_MODEL_DIR=$(hf download pytorch/Phi-4-mini-instruct-INT8-INT4)
72+
EXPECTED_MODEL_SIZE_UPPER_BOUND=$((3 * 1024 * 1024 * 1024)) # 3GB
73+
$PYTHON_EXECUTABLE -m executorch.examples.models.phi_4_mini.convert_weights \
74+
$HF_MODEL_DIR \
75+
pytorch_model_converted.bin
76+
77+
$PYTHON_EXECUTABLE -m executorch.examples.models.llama.export_llama \
78+
--model "phi_4_mini" \
79+
--checkpoint pytorch_model_converted.bin \
80+
--params examples/models/phi_4_mini/config/config.json \
81+
--output_name $MODEL_OUT \
82+
-kv \
83+
--use_sdpa_with_kv_cache \
84+
-X \
85+
--xnnpack-extended-ops \
86+
--max_context_length 1024 \
87+
--max_seq_length 1024 \
88+
--dtype fp32 \
89+
--metadata '{"get_bos_id":199999, "get_eos_ids":[200020,199999]}'
90+
;;
91+
92+
*)
93+
echo "Error: unsupported model_name '$MODEL_NAME'"
94+
echo "Supported values: qwen3_4b, phi_4_mini"
95+
exit 1
96+
;;
97+
esac
98+
99+
# Check file size
100+
MODEL_SIZE=$(stat --printf="%s" $MODEL_OUT 2>/dev/null || stat -f%z $MODEL_OUT)
101+
if [[ $MODEL_SIZE -gt $EXPECTED_MODEL_SIZE_UPPER_BOUND ]]; then
102+
echo "Error: model size $MODEL_SIZE is greater than expected upper bound $EXPECTED_MODEL_SIZE_UPPER_BOUND"
103+
exit 1
104+
fi
105+
106+
# Install ET with CMake
107+
if [[ "$TEST_WITH_RUNNER" -eq 1 ]]; then
108+
echo "[runner] Building and testing llama_main ..."
109+
cmake -DPYTHON_EXECUTABLE=python \
110+
-DCMAKE_INSTALL_PREFIX=cmake-out \
111+
-DEXECUTORCH_ENABLE_LOGGING=1 \
112+
-DCMAKE_BUILD_TYPE=Release \
113+
-DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
114+
-DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
115+
-DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
116+
-DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
117+
-DEXECUTORCH_BUILD_XNNPACK=ON \
118+
-DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
119+
-DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
120+
-DEXECUTORCH_BUILD_EXTENSION_LLM_RUNNER=ON \
121+
-DEXECUTORCH_BUILD_EXTENSION_LLM=ON \
122+
-DEXECUTORCH_BUILD_KERNELS_LLM=ON \
123+
-Bcmake-out .
124+
cmake --build cmake-out -j16 --config Release --target install
125+
126+
127+
# Install llama runner
128+
cmake -DPYTHON_EXECUTABLE=python \
129+
-DCMAKE_BUILD_TYPE=Release \
130+
-Bcmake-out/examples/models/llama \
131+
examples/models/llama
132+
cmake --build cmake-out/examples/models/llama -j16 --config Release
133+
134+
# Run the model
135+
./cmake-out/examples/models/llama/llama_main --model_path=$MODEL_OUT --tokenizer_path="${HF_MODEL_DIR}/tokenizer.json" --prompt="Once upon a time,"
136+
fi
137+
138+
# Clean up
139+
rm -f pytorch_model_converted.bin "$MODEL_OUT"

‎.ci/scripts/unittest-windows.ps1‎

Lines changed: 15 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,15 @@
1+
param (
2+
[string]$editable = $false
3+
)
4+
5+
Set-PSDebug -Trace 1
6+
$ErrorActionPreference = 'Stop'
7+
$PSNativeCommandUseErrorActionPreference = $true
8+
9+
# Run pytest with coverage
10+
# pytest -n auto --cov=./ --cov-report=xml
11+
pytest -v --full-trace -c pytest-windows.ini
12+
if ($LASTEXITCODE -ne 0) {
13+
Write-Host "Pytest invocation was unsuccessful. Exit code: $LASTEXITCODE."
14+
exit $LASTEXITCODE
15+
}

0 commit comments

Comments
 (0)