Skip to content

Commit c22e2f7

Browse files
committed
Update on "[ET-VK] Adding all tensor packing support for native layer norm."
This diff updates Executorch Vulkan backend's `native layer norm` operation to support width, height and channel packed tensors. . and adds new test cases to the cases.py file to test the operation. Differential Revision: [D71663678](https://our.internmc.facebook.com/intern/diff/D71663678/) [ghstack-poisoned]
2 parents 08f893e + 8667235 commit c22e2f7

File tree

493 files changed

+12092
-135344
lines changed

Some content is hidden

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

493 files changed

+12092
-135344
lines changed
Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1 +1 @@
1-
295f2ed4d103017f7e19a7b8263ece606cd629db
1+
7ae0ce6360b6e4f944906502d20da24c04debee5

.ci/scripts/build_android_instrumentation.sh

Lines changed: 6 additions & 26 deletions
Original file line numberDiff line numberDiff line change
@@ -12,30 +12,10 @@ if [[ -z "${PYTHON_EXECUTABLE:-}" ]]; then
1212
fi
1313
which "${PYTHON_EXECUTABLE}"
1414

15-
build_android_test() {
16-
mkdir -p extension/android/executorch_android/src/androidTest/resources
17-
cp extension/module/test/resources/add.pte extension/android/executorch_android/src/androidTest/resources
18-
pushd extension/android
19-
ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:testDebugUnitTest
20-
ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:assembleAndroidTest
21-
popd
22-
}
15+
mkdir -p "${BUILD_AAR_DIR}"/executorch_android/src/androidTest/resources
16+
cp extension/module/test/resources/add.pte "${BUILD_AAR_DIR}"/executorch_android/src/androidTest/resources
2317

24-
collect_artifacts_to_be_uploaded() {
25-
ARTIFACTS_DIR_NAME="$1"
26-
# Collect Java library test
27-
JAVA_LIBRARY_TEST_DIR="${ARTIFACTS_DIR_NAME}/library_test_dir"
28-
mkdir -p "${JAVA_LIBRARY_TEST_DIR}"
29-
cp extension/android/executorch_android/build/outputs/apk/androidTest/debug/*.apk "${JAVA_LIBRARY_TEST_DIR}"
30-
}
31-
32-
main() {
33-
build_android_test
34-
if [ -n "$ARTIFACTS_DIR_NAME" ]; then
35-
collect_artifacts_to_be_uploaded ${ARTIFACTS_DIR_NAME}
36-
fi
37-
}
38-
39-
if [[ "${BASH_SOURCE[0]}" == "${0}" ]]; then
40-
main "$@"
41-
fi
18+
pushd "${BUILD_AAR_DIR}"
19+
ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:testDebugUnitTest
20+
ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:assembleAndroidTest
21+
popd

.ci/scripts/gather_test_models.py

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -14,7 +14,7 @@
1414
from typing import Any
1515

1616
from examples.models import MODEL_NAME_TO_MODEL
17-
from examples.xnnpack import MODEL_NAME_TO_OPTIONS
17+
from examples.xnnpack import MODEL_NAME_TO_OPTIONS, QuantType
1818

1919
DEFAULT_RUNNERS = {
2020
"linux": "linux.2xlarge",
@@ -33,7 +33,7 @@
3333
"dl3": "linux.4xlarge.memory",
3434
"emformer_join": "linux.4xlarge.memory",
3535
"emformer_predict": "linux.4xlarge.memory",
36-
"phi-4-mini": "linux.4xlarge.memory",
36+
"phi_4_mini": "linux.4xlarge.memory",
3737
}
3838
}
3939

@@ -154,7 +154,7 @@ def export_models_for_ci() -> dict[str, dict]:
154154
if backend == "xnnpack":
155155
if name not in MODEL_NAME_TO_OPTIONS:
156156
continue
157-
if MODEL_NAME_TO_OPTIONS[name].quantization:
157+
if MODEL_NAME_TO_OPTIONS[name].quantization != QuantType.NONE:
158158
backend += "-quantization"
159159

160160
if MODEL_NAME_TO_OPTIONS[name].delegation:

.ci/scripts/setup-openvino.sh

Lines changed: 28 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,28 @@
1+
#!/bin/bash
2+
# Copyright (c) Meta Platforms, Inc. and affiliates.
3+
# All rights reserved.
4+
#
5+
# This source code is licensed under the BSD-style license found in the
6+
# LICENSE file in the root directory of this source tree.
7+
8+
set -ex
9+
10+
# shellcheck source=/dev/null
11+
source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
12+
13+
git clone https://github.com/openvinotoolkit/openvino.git
14+
cd openvino && git checkout releases/2025/1
15+
git submodule update --init --recursive
16+
sudo ./install_build_dependencies.sh
17+
mkdir build && cd build
18+
cmake .. -DCMAKE_BUILD_TYPE=Release -DENABLE_PYTHON=ON
19+
make -j$(nproc)
20+
21+
cd ..
22+
cmake --install build --prefix dist
23+
24+
source dist/setupvars.sh
25+
cd ../backends/openvino
26+
pip install -r requirements.txt
27+
cd scripts
28+
./openvino_build.sh --enable_python

.ci/scripts/test_llama.sh

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -269,7 +269,7 @@ $PYTHON_EXECUTABLE -m examples.models.llama.export_llama ${EXPORT_ARGS}
269269

270270
# Create tokenizer.bin.
271271
echo "Creating tokenizer.bin"
272-
$PYTHON_EXECUTABLE -m extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
272+
$PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
273273

274274

275275
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

Lines changed: 1 addition & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -55,7 +55,7 @@ cmake --build cmake-out/examples/models/llama -j16 --config Release
5555
download_stories_model_artifacts
5656

5757
echo "Creating tokenizer.bin"
58-
$PYTHON_EXECUTABLE -m extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
58+
$PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
5959

6060
# Export model
6161
LLAMA_CHECKPOINT=stories110M.pt
@@ -78,7 +78,6 @@ ${PYTHON_EXECUTABLE} -m examples.models.llama.export_llama \
7878
-qmode "torchao:8da${QLINEAR_BITWIDTH}w" \
7979
--group_size ${QLINEAR_GROUP_SIZE} \
8080
-E "torchao:${QEMBEDDING_BITWIDTH},${QEMBEDDING_GROUP_SIZE}" \
81-
--disable_dynamic_shape \
8281
-d fp32
8382

8483
# Test run

.ci/scripts/test_model.sh

Lines changed: 9 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -96,15 +96,15 @@ test_model() {
9696
bash examples/models/llama/install_requirements.sh
9797
# Test export_llama script: python3 -m examples.models.llama.export_llama.
9898
# Use Llama random checkpoint with Qwen 2.5 1.5b model configuration.
99-
"${PYTHON_EXECUTABLE}" -m examples.models.llama.export_llama --model "${MODEL_NAME}" -c examples/models/llama/params/demo_rand_params.pth -p examples/models/qwen2_5/1_5b_config.json
99+
"${PYTHON_EXECUTABLE}" -m examples.models.llama.export_llama --model "${MODEL_NAME}" -p examples/models/qwen2_5/1_5b_config.json
100100
rm "./${MODEL_NAME}.pte"
101101
return # Skip running with portable executor runnner since portable doesn't support Qwen's biased linears.
102102
fi
103-
if [[ "${MODEL_NAME}" == "phi-4-mini" ]]; then
103+
if [[ "${MODEL_NAME}" == "phi_4_mini" ]]; then
104104
# Install requirements for export_llama
105105
bash examples/models/llama/install_requirements.sh
106106
# Test export_llama script: python3 -m examples.models.llama.export_llama.
107-
"${PYTHON_EXECUTABLE}" -m examples.models.llama.export_llama --model "${MODEL_NAME}" -c examples/models/llama/params/demo_rand_params.pth -p examples/models/phi-4-mini/config.json
107+
"${PYTHON_EXECUTABLE}" -m examples.models.llama.export_llama --model "${MODEL_NAME}" -p examples/models/phi_4_mini/config.json
108108
run_portable_executor_runner
109109
rm "./${MODEL_NAME}.pte"
110110
return
@@ -224,19 +224,22 @@ test_model_with_coreml() {
224224

225225
"${PYTHON_EXECUTABLE}" -m examples.apple.coreml.scripts.export --model_name="${MODEL_NAME}" --compute_precision "${DTYPE}"
226226
EXPORTED_MODEL=$(find "." -type f -name "${MODEL_NAME}*.pte" -print -quit)
227-
# TODO:
227+
228228
if [ -n "$EXPORTED_MODEL" ]; then
229229
EXPORTED_MODEL_WITH_DTYPE="${EXPORTED_MODEL%.pte}_${DTYPE}.pte"
230230
mv "$EXPORTED_MODEL" "$EXPORTED_MODEL_WITH_DTYPE"
231231
EXPORTED_MODEL="$EXPORTED_MODEL_WITH_DTYPE"
232-
echo "Renamed file path: $EXPORTED_MODEL"
232+
echo "OK exported model: $EXPORTED_MODEL"
233233
else
234-
echo "No .pte file found"
234+
echo "[error] failed to export model: no .pte file found"
235235
exit 1
236236
fi
237237

238238
# Run the model
239239
if [ "${should_test}" = true ]; then
240+
echo "Installing requirements needed to build coreml_executor_runner..."
241+
backends/apple/coreml/scripts/install_requirements.sh
242+
240243
echo "Testing exported model with coreml_executor_runner..."
241244
local out_dir=$(mktemp -d)
242245
COREML_EXECUTOR_RUNNER_OUT_DIR="${out_dir}" examples/apple/coreml/scripts/build_executor_runner.sh

.ci/scripts/test_openvino.sh

Lines changed: 16 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,16 @@
1+
#!/bin/bash
2+
# Copyright (c) Meta Platforms, Inc. and affiliates.
3+
# All rights reserved.
4+
#
5+
# This source code is licensed under the BSD-style license found in the
6+
# LICENSE file in the root directory of this source tree.
7+
8+
set -ex
9+
10+
# shellcheck source=/dev/null
11+
source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
12+
13+
source openvino/dist/setupvars.sh
14+
cd backends/openvino/tests
15+
python test_runner.py --test_type ops
16+
python test_runner.py --test_type models

.ci/scripts/test_phi_3_mini.sh

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -56,7 +56,7 @@ cmake_build_phi_3_mini() {
5656
prepare_tokenizer() {
5757
echo "Downloading and converting tokenizer.model"
5858
wget -O tokenizer.model "https://huggingface.co/microsoft/Phi-3-mini-128k-instruct/resolve/main/tokenizer.model?download=true"
59-
$PYTHON_EXECUTABLE -m executorch.extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
59+
$PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
6060
}
6161

6262
# Export phi-3-mini model to pte

.ci/scripts/test_qnn_static_llama.sh

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -30,7 +30,7 @@ pip install graphviz
3030
# Download stories llama110m artifacts
3131
download_stories_model_artifacts
3232
echo "Creating tokenizer.bin"
33-
$PYTHON_EXECUTABLE -m extension.llm.tokenizer.tokenizer -t tokenizer.model -o tokenizer.bin
33+
$PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
3434

3535
set +e
3636
# Compile only as weight sharing is not applicable on x86

0 commit comments

Comments
 (0)