Skip to content

GitLab

"...git@developer.sourcefind.cn:OpenDAS/TransformerEngine.git" did not exist on "f8cb598c1c9b6228d39bf77cec3ea27c7ab6a1d7"
Unverified Commit c571c2fd authored by linxiddd's avatar linxiddd Committed by GitHub
Browse files

[QA] Add error handling (#1570) 



* [QA] Add error handling

-Standardize test failure handling using the unified 'test_fail' function and 'error_exit' function.
Signed-off-by: default avatarLinxi Ding <linxid@nvidia.com>

* Update script to use explicit python3, pip3, and python3 -m pytest calls

- Change pip to pip3.
- Change python to python3.
- Change pytest to python3 -m pytest.
Signed-off-by: default avatarLinxi Ding <linxid@nvidia.com>

---------
Signed-off-by: default avatarLinxi Ding <linxid@nvidia.com>
parent 6a855962
Hide whitespace changes
Inline Side-by-side
Showing with 186 additions and 76 deletions
qa/L0_jax_distributed_unittest/test.sh
View file @ c571c2fd
......@@ -2,14 +2,33 @@
#
# See LICENSE for license information.
set -xe
function error_exit() {
echo "Error: $1"
exit 1
}
function test_fail() {
RET=1
FAILED_CASES="$FAILED_CASES $1"
echo "Error: sub-test failed: $1"
}
RET=0
FAILED_CASES=""
: ${TE_PATH:=/opt/transformerengine}
pip3 install -r $TE_PATH/examples/jax/encoder/requirements.txt
pip3 install -r $TE_PATH/examples/jax/encoder/requirements.txt || error_exit "Failed to install requirements"
# Make encoder tests to have run-to-run deterministic to have the stable CI results
export XLA_FLAGS="${XLA_FLAGS} --xla_gpu_deterministic_ops"
python3 -m pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/examples/jax/encoder/test_multigpu_encoder.py
python3 -m pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/examples/jax/encoder/test_model_parallel_encoder.py
. $TE_PATH/examples/jax/encoder/run_test_multiprocessing_encoder.sh
python3 -m pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/examples/jax/encoder/test_multigpu_encoder.py || test_fail "test_multigpu_encoder.py"
python3 -m pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/examples/jax/encoder/test_model_parallel_encoder.py || test_fail "test_model_parallel_encoder.py"
. $TE_PATH/examples/jax/encoder/run_test_multiprocessing_encoder.sh || test_fail "run_test_multiprocessing_encoder.sh"
if [ $RET -ne 0 ]; then
echo "Error: some sub-tests failed: $FAILED_CASES"
exit 1
fi
echo "All tests passed"
exit 0
qa/L0_jax_unittest/test.sh
View file @ c571c2fd
......@@ -2,22 +2,41 @@
#
# See LICENSE for license information.
set -xe
function error_exit() {
echo "Error: $1"
exit 1
}
pip3 install "nltk>=3.8.2"
pip3 install pytest==8.2.1
function test_fail() {
RET=1
FAILED_CASES="$FAILED_CASES $1"
echo "Error: sub-test failed: $1"
}
RET=0
FAILED_CASES=""
pip3 install "nltk>=3.8.2" || error_exit "Failed to install nltk"
pip3 install pytest==8.2.1 || error_exit "Failed to install pytest"
: ${TE_PATH:=/opt/transformerengine}
python3 -m pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/tests/jax -k 'not distributed' --ignore=$TE_PATH/tests/jax/test_praxis_layers.py
python3 -m pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/tests/jax -k 'not distributed' --ignore=$TE_PATH/tests/jax/test_praxis_layers.py || test_fail "test_praxis_layers.py"
# Test without custom calls
NVTE_CUSTOM_CALLS_RE="" python3 -m pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/tests/jax/test_custom_call_compute.py
NVTE_CUSTOM_CALLS_RE="" python3 -m pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/tests/jax/test_custom_call_compute.py || test_fail "test_custom_call_compute.py"
pip3 install -r $TE_PATH/examples/jax/mnist/requirements.txt
pip3 install -r $TE_PATH/examples/jax/encoder/requirements.txt
pip3 install -r $TE_PATH/examples/jax/mnist/requirements.txt || error_exit "Failed to install mnist requirements"
pip3 install -r $TE_PATH/examples/jax/encoder/requirements.txt || error_exit "Failed to install encoder requirements"
python3 -m pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/examples/jax/mnist
python3 -m pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/examples/jax/mnist || test_fail "test_mnist.py"
# Make encoder tests to have run-to-run deterministic to have the stable CI results
export XLA_FLAGS="${XLA_FLAGS} --xla_gpu_deterministic_ops"
python3 -m pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/examples/jax/encoder/test_single_gpu_encoder.py
python3 -m pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/examples/jax/encoder/test_single_gpu_encoder.py || test_fail "test_single_gpu_encoder.py"
if [ $RET -ne 0 ]; then
echo "Error: some sub-tests failed: $FAILED_CASES"
exit 1
fi
echo "All tests passed"
exit 0
qa/L0_jax_wheel/test.sh
View file @ c571c2fd
......@@ -2,34 +2,53 @@
#
# See LICENSE for license information.
set -e
function error_exit() {
echo "Error: $1"
exit 1
}
function test_fail() {
RET=1
FAILED_CASES="$FAILED_CASES $1"
echo "Error: sub-test failed: $1"
}
RET=0
FAILED_CASES=""
: "${TE_PATH:=/opt/transformerengine}"
pip3 install wheel
pip3 install wheel || error_exit "Failed to install wheel"
cd $TE_PATH
pip3 uninstall -y transformer-engine transformer-engine-cu12 transformer-engine-jax
pip3 uninstall -y transformer-engine transformer-engine-cu12 transformer-engine-jax || error_exit "Failed to uninstall transformer-engine transformer-engine-cu12 transformer-engine-jax"
VERSION=`cat $TE_PATH/build_tools/VERSION.txt`
WHL_BASE="transformer_engine-${VERSION}"
# Core wheel.
NVTE_RELEASE_BUILD=1 python3 setup.py bdist_wheel
wheel unpack dist/*
NVTE_RELEASE_BUILD=1 python3 setup.py bdist_wheel || error_exit "Failed to setup bdist_wheel"
wheel unpack dist/* || error_exit "Failed to unpack dist/*"
sed -i "s/Name: transformer-engine/Name: transformer-engine-cu12/g" "transformer_engine-${VERSION}/transformer_engine-${VERSION}.dist-info/METADATA"
sed -i "s/Name: transformer_engine/Name: transformer_engine_cu12/g" "transformer_engine-${VERSION}/transformer_engine-${VERSION}.dist-info/METADATA"
mv "${WHL_BASE}/${WHL_BASE}.dist-info" "${WHL_BASE}/transformer_engine_cu12-${VERSION}.dist-info"
wheel pack ${WHL_BASE}
rm dist/*.whl
mv *.whl dist/
NVTE_RELEASE_BUILD=1 NVTE_BUILD_METAPACKAGE=1 python3 setup.py bdist_wheel
mv "${WHL_BASE}/${WHL_BASE}.dist-info" "${WHL_BASE}/transformer_engine_cu12-${VERSION}.dist-info" || error_exit "Failed to move ${WHL_BASE}.dist-info to transformer_engine_cu12-${VERSION}.dist-info"
wheel pack ${WHL_BASE} || error_exit "Failed to pack ${WHL_BASE}"
rm dist/*.whl || error_exit "Failed to remove dist/*.whl"
mv *.whl dist/ || error_exit "Failed to move *.whl to dist/"
NVTE_RELEASE_BUILD=1 NVTE_BUILD_METAPACKAGE=1 python3 setup.py bdist_wheel || error_exit "Failed to setup metapackage"
cd transformer_engine/jax
NVTE_RELEASE_BUILD=1 python3 setup.py sdist
NVTE_RELEASE_BUILD=1 python3 setup.py sdist || error_exit "Failed to setup sdist"
pip3 install dist/*
pip3 install dist/* || error_exit "Failed to install dist/*"
cd $TE_PATH
pip3 install dist/*.whl --no-deps
pip3 install dist/*.whl --no-deps || error_exit "Failed to install dist/*.whl --no-deps"
python3 $TE_PATH/tests/jax/test_sanity_import.py || test_fail "test_sanity_import.py"
python3 $TE_PATH/tests/jax/test_sanity_import.py
if [ $RET -ne 0 ]; then
echo "Error: some sub-tests failed: $FAILED_CASES"
exit 1
fi
echo "All tests passed"
exit 0
qa/L0_pytorch_unittest/test.sh
View file @ c571c2fd
......@@ -2,29 +2,46 @@
#
# See LICENSE for license information.
function error_exit() {
echo "Error: $1"
exit 1
}
function test_fail() {
RET=1
FAILED_CASES="$FAILED_CASES $1"
echo "Error: sub-test failed: $1"
}
RET=0
FAILED_CASES=""
set -x
: ${TE_PATH:=/opt/transformerengine}
pip3 install pytest==8.2.1
FAIL=0
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_sanity.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_recipe.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_deferred_init.py || FAIL=1
PYTORCH_JIT=0 NVTE_TORCH_COMPILE=0 NVTE_ALLOW_NONDETERMINISTIC_ALGO=0 python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_numerics.py || FAIL=1
NVTE_CUDNN_MXFP8_NORM=0 PYTORCH_JIT=0 NVTE_TORCH_COMPILE=0 NVTE_ALLOW_NONDETERMINISTIC_ALGO=0 python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_cuda_graphs.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_jit.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_fused_rope.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_float8tensor.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_gqa.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_fused_optimizer.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_multi_tensor.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_fusible_ops.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_permutation.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_parallel_cross_entropy.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_cpu_offloading.py || FAIL=1
NVTE_DEBUG=1 NVTE_DEBUG_LEVEL=1 python3 -m pytest -o log_cli=true --log-cli-level=INFO -v -s $TE_PATH/tests/pytorch/fused_attn/test_fused_attn.py || FAIL=1
exit $FAIL
pip3 install pytest==8.2.1 || error_exit "Failed to install pytest"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_sanity.py || test_fail "test_sanity.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_recipe.py || test_fail "test_recipe.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_deferred_init.py || test_fail "test_deferred_init.py"
PYTORCH_JIT=0 NVTE_TORCH_COMPILE=0 NVTE_ALLOW_NONDETERMINISTIC_ALGO=0 python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_numerics.py || test_fail "test_numerics.py"
NVTE_CUDNN_MXFP8_NORM=0 PYTORCH_JIT=0 NVTE_TORCH_COMPILE=0 NVTE_ALLOW_NONDETERMINISTIC_ALGO=0 python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_cuda_graphs.py || test_fail "test_cuda_graphs.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_jit.py || test_fail "test_jit.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_fused_rope.py || test_fail "test_fused_rope.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_float8tensor.py || test_fail "test_float8tensor.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_gqa.py || test_fail "test_gqa.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_fused_optimizer.py || test_fail "test_fused_optimizer.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_multi_tensor.py || test_fail "test_multi_tensor.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_fusible_ops.py || test_fail "test_fusible_ops.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_permutation.py || test_fail "test_permutation.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_parallel_cross_entropy.py || test_fail "test_parallel_cross_entropy.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/test_cpu_offloading.py || test_fail "test_cpu_offloading.py"
NVTE_DEBUG=1 NVTE_DEBUG_LEVEL=1 python3 -m pytest -o log_cli=true --log-cli-level=INFO -v -s $TE_PATH/tests/pytorch/fused_attn/test_fused_attn.py || test_fail "test_fused_attn.py"
if [ "$RET" -ne 0 ]; then
echo "Error in the following test cases:$FAILED_CASES"
exit 1
fi
echo "All tests passed"
exit 0
qa/L0_pytorch_wheel/test.sh
View file @ c571c2fd
......@@ -2,34 +2,53 @@
#
# See LICENSE for license information.
set -e
function error_exit() {
echo "Error: $1"
exit 1
}
function test_fail() {
RET=1
FAILED_CASES="$FAILED_CASES $1"
echo "Error: sub-test failed: $1"
}
RET=0
FAILED_CASES=""
: "${TE_PATH:=/opt/transformerengine}"
pip3 install wheel
pip3 install wheel || error_exit "Failed to install wheel"
cd $TE_PATH
pip3 uninstall -y transformer-engine transformer-engine-cu12 transformer-engine-torch
pip3 uninstall -y transformer-engine transformer-engine-cu12 transformer-engine-torch || error_exit "Failed to uninstall transformer-engine transformer-engine-cu12 transformer-engine-torch"
VERSION=`cat $TE_PATH/build_tools/VERSION.txt`
WHL_BASE="transformer_engine-${VERSION}"
# Core wheel.
NVTE_RELEASE_BUILD=1 python3 setup.py bdist_wheel
wheel unpack dist/*
NVTE_RELEASE_BUILD=1 python3 setup.py bdist_wheel || error_exit "Failed to setup bdist_wheel"
wheel unpack dist/* || error_exit "Failed to unpack dist/*"
sed -i "s/Name: transformer-engine/Name: transformer-engine-cu12/g" "transformer_engine-${VERSION}/transformer_engine-${VERSION}.dist-info/METADATA"
sed -i "s/Name: transformer_engine/Name: transformer_engine_cu12/g" "transformer_engine-${VERSION}/transformer_engine-${VERSION}.dist-info/METADATA"
mv "${WHL_BASE}/${WHL_BASE}.dist-info" "${WHL_BASE}/transformer_engine_cu12-${VERSION}.dist-info"
wheel pack ${WHL_BASE}
rm dist/*.whl
mv *.whl dist/
NVTE_RELEASE_BUILD=1 NVTE_BUILD_METAPACKAGE=1 python3 setup.py bdist_wheel
mv "${WHL_BASE}/${WHL_BASE}.dist-info" "${WHL_BASE}/transformer_engine_cu12-${VERSION}.dist-info" || error_exit "Failed to move ${WHL_BASE}.dist-info to transformer_engine_cu12-${VERSION}.dist-info"
wheel pack ${WHL_BASE} || error_exit "Failed to pack ${WHL_BASE}"
rm dist/*.whl || error_exit "Failed to remove dist/*.whl"
mv *.whl dist/ || error_exit "Failed to move *.whl to dist/"
NVTE_RELEASE_BUILD=1 NVTE_BUILD_METAPACKAGE=1 python3 setup.py bdist_wheel || error_exit "Failed to setup metapackage"
cd transformer_engine/pytorch
NVTE_RELEASE_BUILD=1 python3 setup.py sdist
NVTE_RELEASE_BUILD=1 python3 setup.py sdist || error_exit "Failed to setup sdist"
pip3 install dist/*
pip3 install dist/* || error_exit "Failed to install dist/*"
cd $TE_PATH
pip3 install dist/*.whl --no-deps
pip3 install dist/*.whl --no-deps || error_exit "Failed to install dist/*.whl --no-deps"
python3 $TE_PATH/tests/pytorch/test_sanity_import.py || test_fail "test_sanity_import.py"
python3 $TE_PATH/tests/pytorch/test_sanity_import.py
if [ "$RET" -ne 0 ]; then
echo "Error in the following test cases:$FAILED_CASES"
exit 1
fi
echo "All tests passed"
exit 0
qa/L1_jax_distributed_unittest/test.sh
View file @ c571c2fd
......@@ -6,4 +6,4 @@ set -xe
: ${TE_PATH:=/opt/transformerengine}
pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/tests/jax/test_distributed_*
python3 -m pytest -c $TE_PATH/tests/jax/pytest.ini -v $TE_PATH/tests/jax/test_distributed_*
qa/L1_pytorch_distributed_unittest/test.sh
View file @ c571c2fd
......@@ -2,17 +2,34 @@
#
# See LICENSE for license information.
: ${TE_PATH:=/opt/transformerengine}
function error_exit() {
echo "Error: $1"
exit 1
}
function test_fail() {
RET=1
FAILED_CASES="$FAILED_CASES $1"
echo "Error: sub-test failed: $1"
}
pip3 install pytest==8.2.1
RET=0
FAILED_CASES=""
: ${TE_PATH:=/opt/transformerengine}
FAIL=0
pip3 install pytest==8.2.1 || error_exit "Failed to install pytest"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/distributed/test_numerics.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/distributed/test_fusible_ops.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/distributed/test_torch_fsdp2.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/distributed/test_comm_gemm_overlap.py || FAIL=1
# python3 -m pytest -v -s $TE_PATH/tests/pytorch/distributed/test_fusible_ops_with_userbuffers.py || FAIL=1 ### TODO Debug UB support with te.Sequential
python3 -m pytest -v -s $TE_PATH/tests/pytorch/fused_attn/test_fused_attn_with_cp.py || FAIL=1
python3 -m pytest -v -s $TE_PATH/tests/pytorch/distributed/test_numerics.py || test_fail "test_numerics.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/distributed/test_fusible_ops.py || test_fail "test_fusible_ops.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/distributed/test_torch_fsdp2.py || test_fail "test_torch_fsdp2.py"
python3 -m pytest -v -s $TE_PATH/tests/pytorch/distributed/test_comm_gemm_overlap.py || test_fail "test_comm_gemm_overlap.py"
# python3 -m pytest -v -s $TE_PATH/tests/pytorch/distributed/test_fusible_ops_with_userbuffers.py || test_fail "test_fusible_ops_with_userbuffers.py" ### TODO Debug UB support with te.Sequential
python3 -m pytest -v -s $TE_PATH/tests/pytorch/fused_attn/test_fused_attn_with_cp.py || test_fail "test_fused_attn_with_cp.py"
exit $FAIL
if [ "$RET" -ne 0 ]; then
echo "Error in the following test cases:$FAILED_CASES"
exit 1
fi
echo "All tests passed"
exit 0
qa/L3_pytorch_FA_versions_test/test.sh
View file @ c571c2fd
......@@ -35,6 +35,6 @@ do
fi
# Run tests
NVTE_TORCH_COMPILE=0 pytest -v -s $TE_PATH/tests/pytorch/fused_attn/test_fused_attn.py
NVTE_TORCH_COMPILE=0 python3 -m pytest -v -s $TE_PATH/tests/pytorch/fused_attn/test_fused_attn.py
done
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment