Error on import protable_lib via pybindings

[guangy10]

🐛 Describe the bug

Here is the error and stack trace:

Traceback:
../miniconda3/envs/executorch/lib/python3.11/importlib/__init__.py:126: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
tests/models/test_modeling_pvt.py:23: in <module>
    from executorch.extension.pybindings.portable_lib import ExecuTorchModule
../miniconda3/envs/executorch/lib/python3.11/site-packages/executorch/extension/pybindings/portable_lib.py:36: in <module>
    from executorch.extension.pybindings._portable_lib import (  # noqa: F401
E   ImportError: dlopen(/Users/guangyang/miniconda3/envs/executorch/lib/python3.11/site-packages/executorch/extension/pybindings/_portable_lib.cpython-311-darwin.so, 0x0002): symbol not found in flat namespace '__ZN10executorch7runtime8internal18resize_tensor_implEPN5torch8executor10TensorImplENS0_8ArrayRefIiEE'

I have confirmed this issue is new and exist on both release/0.6 and main (7030b33).

I notice this issue when updating my local checkout to latest main, as a comparson, the pybindings works fine on an older commit (0342babc505bcb90244874e9ed9218d90dd67b87) from Tue this week, so likely the issue is introduced very recently.

I caught this issue on optimum-executch, unfortunately, I don't know if there is another way to repo, I will just share the repo steps from optimum-executch:

1. clone the optimum-executorch repo and follow the setup guide here: https://github.com/huggingface/optimum-executorch
2. run an unit test: RUN_SLOW=1 pytest tests/models/test_modeling_bert.py -v -s -k mask
   You will see the error

Versions

Collecting environment information...
PyTorch version: 2.7.0.dev20250310
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: macOS 15.3.2 (arm64)
GCC version: Could not collect
Clang version: 16.0.0 (clang-1600.0.26.6)
CMake version: version 3.31.4
Libc version: N/A

Python version: 3.11.11 (main, Dec 11 2024, 10:25:04) [Clang 14.0.6 ] (64-bit runtime)
Python platform: macOS-15.3.2-arm64-arm-64bit
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Apple M1 Max

Versions of relevant libraries:
[pip3] executorch==0.6.0a0+92408be
[pip3] flake8==6.1.0
[pip3] flake8-breakpoint==1.1.0
[pip3] flake8-bugbear==24.4.26
[pip3] flake8-comprehensions==3.14.0
[pip3] flake8-plugin-utils==1.3.3
[pip3] flake8-pyi==23.5.0
[pip3] mypy==1.14.1
[pip3] mypy-extensions==1.0.0
[pip3] numpy==2.2.4
[pip3] optimum-executorch==0.0.0.dev0
[pip3] torch==2.7.0.dev20250310
[pip3] torchao==0.10.0+git923242e2
[pip3] torchaudio==2.6.0.dev20250310
[pip3] torchsr==1.0.4
[pip3] torchvision==0.22.0.dev20250310
[conda] executorch 0.6.0a0+92408be pypi_0 pypi
[conda] numpy 2.2.4 pypi_0 pypi
[conda] optimum-executorch 0.0.0.dev0 pypi_0 pypi
[conda] torch 2.7.0.dev20250310 pypi_0 pypi
[conda] torchao 0.10.0+git923242e2 pypi_0 pypi
[conda] torchaudio 2.6.0.dev20250310 pypi_0 pypi
[conda] torchfix 0.6.0 pypi_0 pypi
[conda] torchsr 1.0.4 pypi_0 pypi
[conda] torchvision 0.22.0.dev20250310 pypi_0 pypi

cc @mergennachin @byjlw

[guangy10]

Max and I are digging into this issue a log. We tried creating new conda env and inspecting the installed pybindings under the correct conda env, and we also tried completely deleting optimum-executorch locally and make a fresh clone and installation. None makes any difference to us. That is, things keep working for him, and keeps failing for me.

Finally I can get things working after I deleted executorch completely, and re-clone and install it from scratch (within an existing conda env, that is, everything else remains unchanged). And unfortunately, we lost all hints to debug into this weird issue after removing the local copy.

Just a heads-up here, if anyone recently run into the same issue, please let us know, per Max's suggestion, we may want to try more things, it could be the pycahce causing the issue, or we may need to provide a way to uninstall first.

[guangy10]

For now, it should not be a blocker for 0.6 release, so I will move it back to our backlog then. cc: @mergennachin

[guangy10]

tagged the task

[larryliu0820]

@guangy10 does this still repro?

[DerrickJ1612]

I think this may be the same error I'm seeing. I'm building from source targeting a RubikPi.
I'm able to do the following:
exported_program = torch.export.export(pose_model, (example_input,))
executorch_program = to_edge_transform_and_lower( exported_program, partitioner = [XnnpackPartitioner()] ).to_executorch()
with open("pose_hrnet_w48_xnnpack_executorch.pte", "wb") as f: f.write(executorch_program.buffer)

When I try
from executorch.runtime import Runtime

OUTPUT

Successfully installed executorch-0.6.0a0+65ab993
(env_executorch) ubuntu@rubikpi:~/Desktop/git/executorch$ python -c "import executorch"
(env_executorch) ubuntu@rubikpi:~/Desktop/git/executorch$ python -c "from executorch.runtime import Runtime"
/home/ubuntu/Desktop/git/executorch/exir/dialects/edge/_ops.py:9: UserWarning: Module executorch was already imported from None, but /home/ubuntu/Desktop/git/executorch is being added to sys.path
import pkg_resources
Traceback (most recent call last):
File "/home/ubuntu/Desktop/git/executorch/runtime/init.py", line 48, in
from executorch.extension.pybindings.portable_lib import (
File "/home/ubuntu/Desktop/git/executorch/extension/pybindings/portable_lib.py", line 36, in
from executorch.extension.pybindings._portable_lib import ( # noqa: F401
ModuleNotFoundError: No module named 'executorch.extension.pybindings._portable_lib'

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
File "", line 1, in
File "/home/ubuntu/Desktop/git/executorch/runtime/init.py", line 54, in
raise ModuleNotFoundError(
ModuleNotFoundError: Prebuilt /extension/pybindings/_portable_lib.so is not found. Please reinstall ExecuTorch from pip.

I do not see this .so file in the directory, I did try to do a minimal build targeting execution runner
cmake .. -DPYTHON_EXECUTABLE=$(which python) -DCMAKE_BUILD_TYPE=Release -DEXECUTORCH_BUILD_PYBIND=OFF -DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_XNNPACK=OFF -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=OFF -DCMAKE_POLICY_VERSION_MINIMUM=3.5

This didn't help either. Any ideas?