Skip to content

Commit 4de2d60

Browse files
authored
Merge branch 'main' into ykhrustalev/cross-compile-sleef
2 parents 2f574bf + 17047ea commit 4de2d60

File tree

342 files changed

+10081
-3504
lines changed

Some content is hidden

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

342 files changed

+10081
-3504
lines changed

.ci/scripts/test_huggingface_optimum_model.py

Lines changed: 9 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -262,14 +262,20 @@ def test_vit(model_id, model_dir, recipe, *, quantize=False, run_only=False):
262262

263263
assert torch.allclose(
264264
eager_output.logits, et_output, atol=1e-02, rtol=1e-02
265-
), "CoreML output does not match eager"
265+
), "Model output does not match eager"
266266

267267

268268
if __name__ == "__main__":
269269
parser = argparse.ArgumentParser()
270270
parser.add_argument("--model", type=str, required=True)
271271
parser.add_argument("--recipe", type=str, required=True)
272272
parser.add_argument("--quantize", action="store_true", help="Enable quantization")
273+
parser.add_argument(
274+
"--model_dir",
275+
type=str,
276+
required=False,
277+
help="When provided, write the pte file to this directory. Otherwise, a temporary directory is created for the test.",
278+
)
273279
args = parser.parse_args()
274280

275281
model_to_model_id_and_test_function = {
@@ -294,11 +300,11 @@ def test_vit(model_id, model_dir, recipe, *, quantize=False, run_only=False):
294300
f"Unknown model name: {args.model}. Available models: {model_to_model_id_and_test_function.keys()}"
295301
)
296302

303+
model_id, test_fn = model_to_model_id_and_test_function[args.model]
297304
with tempfile.TemporaryDirectory() as tmp_dir:
298-
model_id, test_fn = model_to_model_id_and_test_function[args.model]
299305
test_fn(
300306
model_id=model_id,
301-
model_dir=tmp_dir,
307+
model_dir=tmp_dir if args.model_dir is None else args.model_dir,
302308
recipe=args.recipe,
303309
quantize=args.quantize,
304310
)

.ci/scripts/test_llama_lora.sh

Lines changed: 51 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -48,8 +48,17 @@ DOWNLOADED_PATH=$(
4848
--model_id "${HF_MODEL_REPO}" \
4949
--files "adapter_config.json" "adapter_model.pt" "consolidated.00.pth" "params.json" "tokenizer.model"
5050
)
51-
EXPORTED_MODEL_NAME="llama_3_2_1B_lora.pte"
52-
# Export model.
51+
# Build llama runner.
52+
cmake_install_executorch_libraries
53+
cmake_build_llama_runner
54+
55+
# Constants.
56+
RUNTIME_ARGS="--tokenizer_path=${DOWNLOADED_PATH}/tokenizer.model --temperature=0 --seq_len=20 --warmup=1"
57+
PROMPT="What happens if you eat watermelon seeds?"
58+
EXPECTED_PREFIX="What happens if you eat watermelon seeds? Watermelon seeds are a good source of vitamin C,"
59+
60+
# Export LoRA PTE file.
61+
MODEL_NAME="llama_3_2_1B_lora"
5362
$PYTHON_EXECUTABLE -m extension.llm.export.export_llm \
5463
base.checkpoint="${DOWNLOADED_PATH}/consolidated.00.pth" \
5564
base.params="${DOWNLOADED_PATH}/params.json" \
@@ -61,36 +70,64 @@ $PYTHON_EXECUTABLE -m extension.llm.export.export_llm \
6170
model.dtype_override="fp32" \
6271
backend.xnnpack.enabled=true \
6372
backend.xnnpack.extended_ops=true \
64-
export.output_name="${EXPORTED_MODEL_NAME}"
65-
66-
# Build llama runner.
67-
cmake_install_executorch_libraries
68-
cmake_build_llama_runner
73+
export.output_name="${MODEL_NAME}.pte"
6974

70-
PROMPT="What happens if you eat watermelon seeds?"
7175
# Run llama runner
72-
RUNTIME_ARGS="--model_path=${EXPORTED_MODEL_NAME} --tokenizer_path=${DOWNLOADED_PATH}/tokenizer.model --temperature=0 --seq_len=20 --warmup=1"
73-
7476
NOW=$(date +"%H:%M:%S")
7577
echo "Starting to run llama runner at ${NOW}"
7678
# shellcheck source=/dev/null
77-
cmake-out/examples/models/llama/llama_main --prompt="${PROMPT}" ${RUNTIME_ARGS} > result.txt
79+
cmake-out/examples/models/llama/llama_main --model_path=${MODEL_NAME}.pte --prompt="${PROMPT}" ${RUNTIME_ARGS} > result.txt
7880
NOW=$(date +"%H:%M:%S")
7981
echo "Finished at ${NOW}"
8082

8183
RESULT=$(cat result.txt)
82-
EXPECTED_PREFIX="What happens if you eat watermelon seeds? Watermelon seeds are a good source of vitamin C,"
83-
8484
if [[ "${RESULT}" == "${EXPECTED_PREFIX}"* ]]; then
8585
echo "Expected result prefix: ${EXPECTED_PREFIX}"
8686
echo "Actual result: ${RESULT}"
87+
# Do not clean up files if test passes, as they're re-used in the next test.
8788
echo "Success"
88-
cleanup_files
8989
else
9090
echo "Expected result prefix: ${EXPECTED_PREFIX}"
9191
echo "Actual result: ${RESULT}"
9292
echo "Failure; results not the same"
93+
cleanup_files
94+
exit 1
95+
fi
9396

97+
# Export LoRA PTE, PTD file.
98+
MODEL_SEPARATE="${MODEL_NAME}_separate"
99+
$PYTHON_EXECUTABLE -m extension.llm.export.export_llm \
100+
base.checkpoint="${DOWNLOADED_PATH}/consolidated.00.pth" \
101+
base.params="${DOWNLOADED_PATH}/params.json" \
102+
base.adapter_checkpoint="${DOWNLOADED_PATH}/adapter_model.pt" \
103+
base.adapter_config="${DOWNLOADED_PATH}/adapter_config.json" \
104+
base.tokenizer_path="${DOWNLOADED_PATH}/tokenizer.model" \
105+
model.use_kv_cache=true \
106+
model.use_sdpa_with_kv_cache=true \
107+
model.dtype_override="fp32" \
108+
backend.xnnpack.enabled=true \
109+
backend.xnnpack.extended_ops=true \
110+
export.output_name="${MODEL_SEPARATE}.pte" \
111+
export.foundation_weights_file="${MODEL_SEPARATE}.ptd"
112+
113+
# Run llama runner.
114+
NOW=$(date +"%H:%M:%S")
115+
echo "Starting to run llama runner at ${NOW}"
116+
# shellcheck source=/dev/null
117+
cmake-out/examples/models/llama/llama_main --model_path=${MODEL_SEPARATE}.pte --data_path=${MODEL_SEPARATE}.ptd --prompt="${PROMPT}" ${RUNTIME_ARGS} > result2.txt
118+
NOW=$(date +"%H:%M:%S")
119+
echo "Finished at ${NOW}"
120+
121+
RESULT2=$(cat result2.txt)
122+
if [[ "${RESULT2}" == "${EXPECTED_PREFIX}"* ]]; then
123+
echo "Expected result prefix: ${EXPECTED_PREFIX}"
124+
echo "Actual result: ${RESULT2}"
125+
echo "Success"
126+
cleanup_files
127+
else
128+
echo "Expected result prefix: ${EXPECTED_PREFIX}"
129+
echo "Actual result: ${RESULT2}"
130+
echo "Failure; results not the same"
94131
cleanup_files
95132
exit 1
96133
fi

.ci/scripts/test_model.sh

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -199,6 +199,8 @@ test_model_with_qnn() {
199199
EXPORT_SCRIPT=albert
200200
elif [[ "${MODEL_NAME}" == "bert" ]]; then
201201
EXPORT_SCRIPT=bert
202+
elif [[ "${MODEL_NAME}" == "conv_former" ]]; then
203+
EXPORT_SCRIPT=conv_former
202204
elif [[ "${MODEL_NAME}" == "cvt" ]]; then
203205
EXPORT_SCRIPT=cvt
204206
elif [[ "${MODEL_NAME}" == "distilbert" ]]; then
@@ -238,7 +240,7 @@ test_model_with_qnn() {
238240
"cvt"|"dit"|"focalnet"|"mobilevit_v2"|"pvt"|"swin")
239241
SCRIPT_FOLDER=oss_scripts
240242
;;
241-
"albert"|"bert"|"distilbert"|"roberta"|"efficientnet"|"mobilevit_v1")
243+
"albert"|"bert"|"conv_former"|"distilbert"|"roberta"|"efficientnet"|"mobilevit_v1")
242244
pip install evaluate
243245
SCRIPT_FOLDER=oss_scripts
244246
# 16bit models will encounter op validation fail on some operations,

.ci/scripts/test_qnn_static_llama.sh

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -33,12 +33,12 @@ echo "Creating tokenizer.bin"
3333
$PYTHON_EXECUTABLE -m pytorch_tokenizers.tools.llama2c.convert -t tokenizer.model -o tokenizer.bin
3434

3535
set +e
36-
# Compile only as weight sharing is not applicable on x86
37-
$PYTHON_EXECUTABLE backends/qualcomm/tests/test_qnn_delegate.py -k TestExampleLLMScript.test_llama_stories_110m --model SM8650 --build_folder build-android/ --executorch_root . --artifact_dir . --llama_artifacts . --compile_only
36+
# Compile only as weight sharing is not applicable on x86.
37+
$PYTHON_EXECUTABLE backends/qualcomm/tests/test_qnn_delegate.py -k TestExampleLLMScript.test_llama_stories_110m --model SM8650 --build_folder build-android/ --executorch_root . --artifact_dir ./stories_110m_pte_size --llama_artifacts . --compile_only
3838
exit_code1=$?
3939

4040
# Checks accuracy with weight sharing disabled since x86 does not support weight sharing.
41-
$PYTHON_EXECUTABLE backends/qualcomm/tests/test_qnn_delegate.py -k TestExampleLLMScript.test_llama_stories_110m --model SM8650 --build_folder build-x86/ --executorch_root . --artifact_dir . --llama_artifacts . --enable_x86_64
41+
$PYTHON_EXECUTABLE backends/qualcomm/tests/test_qnn_delegate.py -k TestExampleLLMScript.test_llama_stories_110m --model SM8650 --build_folder build-x86/ --executorch_root . --artifact_dir ./stories_110m_accuracy --llama_artifacts . --enable_x86_64
4242
exit_code2=$?
4343

4444
# Check BC
Lines changed: 93 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,93 @@
1+
name: Add Open External Contributor PRs and Issues to PyTorch Org Project 136
2+
3+
on:
4+
schedule:
5+
- cron: '0 * * * *'
6+
workflow_dispatch:
7+
8+
jobs:
9+
add_to_project:
10+
runs-on: ubuntu-latest
11+
steps:
12+
- name: Add open issues and open, non-draft PRs to org project (excluding certain authors)
13+
uses: actions/github-script@v7
14+
with:
15+
github-token: ${{ secrets.PYTORCH_PROJECT_PAT }}
16+
script: |
17+
const projectId = "PVT_kwDOAUB9vs4A_PUL"; // PyTorch org project 136
18+
const owner = 'pytorch';
19+
const repo = 'executorch';
20+
21+
// List of authors to exclude
22+
const excludedAuthors = new Set([
23+
"nil-is-all", "cbilgin", "KimishPatel", "psiddh", "digantdesai", "SS-JIA", "ahmtox", "mcr229", "shoumikhin",
24+
"manuelcandales", "metascroy", "cccclai", "rohansjoshi", "kirklandsign", "abhinaykukkadapu", "JacobSzwejbka",
25+
"Conarnar", "lucylq", "larryliu0820", "BujSet", "Gasoonjia", "Juntian777", "guangy10", "jackzhxng",
26+
"GregoryComer", "leafs1", "swolchok", "mergennachin", "tarun292", "byjlw", "jathu", "Jack-Khuu", "georgehong",
27+
"zhenyan-zhang-meta", "silverguo", "dbort", "jorgep31415", "huydhn", "mcremon-meta", "trivedivivek", "angelayi",
28+
"helunwencser", "hsharma35", "zhxchen17", "iseeyuan", "svekars", "nathanaelsee", "dulinriley", "jerryzh168",
29+
"cmodi-meta", "bigfootjon", "sxu", "ydwu4", "Riandy", "tugsbayasgalan", "bsoyluoglu", "yangw-dev", "YIWENX14",
30+
"namanahuja", "yushangdi", "limintang", "pianpwk", "viveknayakatmeta", "andreanicastro", "JakeStevens",
31+
"gmagogsfm", "zonglinpeng", "eigen-k", "derekxu", "salilsdesai", "skrtskrtfb", "pssrawat", "r-barnes", "pytorchbot",
32+
"pytorchmergebot", "pytorchupdatebot", "facebook-github-bot", "Erik-Lundell", "zingo", "AdrianLundell",
33+
"oscarandersson8218", "per", "Sebastian-Larsson", "SaoirseARM", "robell", "mansnils", "martinlsm", "freddan80",
34+
"YufengShi-dudu", "tom-arm", "perheld", "Jerry-Ge", "gggekov", "fumchin", "wwwind", "haowhsu-quic", "shewu-quic",
35+
"winskuo-quic", "chunit-quic", "DannyYuyang-quic", "chuntl", "cymbalrush", "DenisVieriu97", "billmguo",
36+
"StrycekSimon", "jirioc", "robert-kalmar", "skywall", "neuropilot-captain"
37+
]);
38+
39+
async function addItem(contentId, type, number) {
40+
try {
41+
await github.graphql(`
42+
mutation {
43+
addProjectV2ItemById(input: {projectId: "${projectId}", contentId: "${contentId}"}) {
44+
item { id }
45+
}
46+
}
47+
`);
48+
console.log(`Added ${type} #${number} to project`);
49+
} catch (error) {
50+
if (error.message && error.message.includes("A project item already exists for this content")) {
51+
// Ignore if already exists
52+
console.log(`${type} #${number} already in project`);
53+
} else {
54+
console.log(`Error adding ${type} #${number}: ${error.message}`);
55+
}
56+
}
57+
}
58+
59+
try {
60+
// Add open issues (not PRs) and exclude by author
61+
const issues = await github.paginate(
62+
github.rest.issues.listForRepo,
63+
{
64+
owner,
65+
repo,
66+
state: 'open',
67+
filter: 'all'
68+
}
69+
);
70+
for (const issue of issues) {
71+
if (!issue.pull_request && !excludedAuthors.has(issue.user.login)) {
72+
await addItem(issue.node_id, 'issue', issue.number);
73+
}
74+
}
75+
76+
// Add open, non-draft PRs (regardless of review state), exclude by author
77+
const prs = await github.paginate(
78+
github.rest.pulls.list,
79+
{
80+
owner,
81+
repo,
82+
state: 'open',
83+
draft: false,
84+
}
85+
);
86+
for (const pr of prs) {
87+
if (!excludedAuthors.has(pr.user.login)) {
88+
await addItem(pr.node_id, 'pr', pr.number);
89+
}
90+
}
91+
} catch (error) {
92+
core.setFailed(`Workflow failed: ${error.message}`);
93+
}

.github/workflows/build-presets.yml

Lines changed: 0 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -6,8 +6,6 @@ on:
66
branches:
77
- main
88
- release/*
9-
paths:
10-
- .github/workflows/build-presets.yml
119
workflow_dispatch:
1210

1311
concurrency:

.github/workflows/pull.yml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -315,7 +315,7 @@ jobs:
315315
bash examples/models/moshi/mimi/install_requirements.sh
316316
317317
# reinstall executorch
318-
bash ./install_executorch.sh
318+
bash ./install_executorch.sh --minimal
319319
320320
# run python unittest
321321
python -m unittest examples.models.moshi.mimi.test_mimi

.github/workflows/trunk.yml

Lines changed: 63 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -60,7 +60,7 @@ jobs:
6060
uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
6161
strategy:
6262
matrix:
63-
model: [add]
63+
model: [add, softmax, mv2]
6464
fail-fast: false
6565
with:
6666
runner: linux.2xlarge
@@ -72,31 +72,85 @@ jobs:
7272
MODEL_NAME=${{ matrix.model }}
7373
CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
7474
conda activate "${CONDA_ENV}"
75+
if [[ ${{ matrix.model}} == "add" ]]; then
76+
SIM_LIMIT_SEC=60
77+
elif [[ ${{ matrix.model}} == "softmax" ]]; then
78+
SIM_LIMIT_SEC=60
79+
elif [[ ${{ matrix.model}} == "mv2" ]]; then
80+
SIM_LIMIT_SEC=5000
81+
else
82+
echo "Failed unsupported model selection ${{ matrix.model }}"
83+
exit 1
84+
fi
7585
7686
source .ci/scripts/utils.sh
7787
source .ci/scripts/zephyr-utils.sh
7888
mkdir -p zephyr_scratch/
7989
cd zephyr_scratch
8090
export ZEPHYR_PROJ_ROOT=$(realpath $(pwd))
91+
export ARM_FVP_TUTORIALS_ROOT=$ZEPHYR_PROJ_ROOT/zephyr/samples/modules/executorch/arm-fvp-tutorials
8192
93+
# TODO @Bujji: Should see if this can be moved into the docker image itself
8294
download_arm_zephyr_sdk
8395
./zephyr-sdk-0.16.0/setup.sh -c -t arm-zephyr-eabi
84-
8596
cd $ZEPHYR_PROJ_ROOT
8697
setup_zephyr_et_module
8798
99+
# Run setup scripts for Arm FVP and Arm AOT Compilation
88100
cd $ZEPHYR_PROJ_ROOT/modules/lib/executorch
89101
install_executorch "--use-pt-pinned-commit"
90102
.ci/scripts/setup-arm-baremetal-tools.sh --target-toolchain zephyr
91103
source examples/arm/ethos-u-scratch/setup_path.sh
92104
source $ZEPHYR_PROJ_ROOT/zephyr/zephyr-env.sh
93-
cd $ZEPHYR_PROJ_ROOT/zephyr/samples/modules/executorch/arm/hello_world
94-
west build -p always -b mps3/corstone300/fvp
95-
FVP_Corstone_SSE-300_Ethos-U55 -a build/zephyr/zephyr.elf -C mps3_board.visualisation.disable-visualisation=1 -C mps3_board.telnetterminal0.start_telnet=0 -C mps3_board.uart0.out_file='sim.out' -C cpu0.CFGITCMSZ=15 -C cpu0.CFGDTCMSZ=15 --simlimit 120
96105
97-
grep -qF "Output[0][0]: (float) 2.000000" sim.out
106+
# Get the model as PTE
107+
python -m examples.arm.aot_arm_compiler \
108+
--model_name="${MODEL_NAME}" \
109+
--output="${MODEL_NAME}.pte"
110+
111+
# Generate the C-style header
112+
cd $ARM_FVP_TUTORIALS_ROOT
113+
python build_model.py \
114+
--executorch-root $ZEPHYR_PROJ_ROOT/modules/lib/executorch \
115+
--pte-file $ZEPHYR_PROJ_ROOT/modules/lib/executorch/${MODEL_NAME}.pte \
116+
--output-path $ARM_FVP_TUTORIALS_ROOT/models/${MODEL_NAME}/src/
117+
118+
cd $ARM_FVP_TUTORIALS_ROOT/models/${MODEL_NAME}/
119+
120+
# Build the zephyr elf
121+
west build -p always -b mps3/corstone300/fvp -- \
122+
-DET_PTE_FILE_PATH_FOR_SELECTIVE_BUILD=$ZEPHYR_PROJ_ROOT/modules/lib/executorch/${MODEL_NAME}.pte
123+
124+
# Run the simulation
125+
FVP_Corstone_SSE-300_Ethos-U55 -a build/zephyr/zephyr.elf \
126+
-C mps3_board.visualisation.disable-visualisation=1 \
127+
-C mps3_board.telnetterminal0.start_telnet=0 \
128+
-C mps3_board.uart0.out_file='sim.out' \
129+
-C cpu0.CFGITCMSZ=15 \
130+
-C cpu0.CFGDTCMSZ=15 \
131+
--simlimit ${SIM_LIMIT_SEC}
132+
133+
# Disable exit on error
134+
set +e
135+
# Report failure if any of the ouptut verification checks fail
136+
grep -qF "ERROR" sim.out
137+
exit_status=$? #store 0 if found (failure), 1 if not (success)
138+
if [[ "$exit_status" -eq "0" ]]; then
139+
cat sim.out
140+
set -e
141+
exit 1
142+
fi
143+
144+
# Report fail if simulation does not complete successfully
145+
grep -qF "SUCCESS: Program complete, exiting." sim.out
98146
exit_status=$? #store 0 if found (success), 1 if not (failure)
99-
exit $exit_status
147+
if [[ "$exit_status" -eq "1" ]]; then
148+
cat sim.out
149+
set -e
150+
exit 1
151+
fi
152+
# Re-enable exit on error
153+
set -e
100154
101155
test-models-linux-aarch64:
102156
name: test-models-linux-aarch64
@@ -234,6 +288,7 @@ jobs:
234288
- test_arm_baremetal: test_models_tosa
235289
- test_arm_baremetal: test_models_ethos-u55
236290
- test_arm_baremetal: test_models_ethos-u85
291+
- test_arm_baremetal: test_smaller_stories_llama
237292
fail-fast: false
238293
with:
239294
runner: linux.2xlarge.memory
@@ -568,7 +623,7 @@ jobs:
568623
strategy:
569624
matrix:
570625
dtype: [fp32]
571-
model: [dl3, mv3, mv2, ic4, ic3, vit, mb, w2l]
626+
model: [dl3, mv3, mv2, ic4, ic3, vit, mb, w2l, conv_former]
572627
fail-fast: false
573628
with:
574629
runner: linux.2xlarge

0 commit comments

Comments
 (0)