OneComp with Llama.cpp conflicts OpenMP execution between PyTorch and Llama.cpp on macOS

[y-vectorfield]

Summary

When we execute OneComp with Llama.cpp on macOS, it conflicts OpenMP execution between PyTorch and Llama.cpp.

Environment

• macOS Tahoe 26.6.1
• CPU: M1 Pro
• uv 0.11.17
• Python 3.12.14
• OneComp 1.13.1
• torch 2.13.0
• llama-cpp-python 0.3.34

What Happened

• If you have installed gcc or clang on macOS using a tool like Homebrew, OpenMP associated with them will conflict with OpenMP within the .venv (generated by uv) directory.
• OneComp's PyTorch backend is configured to use OpenMP library located within .venv.
• Depending on the environment, Llama.cpp is configured to use the OpenMP library associated with gcc or clang.
• Therefore, the above two will result in a conflict at runtime.

If the following warning message is shown, this conflict may be occurring.

**/multiprocessing/resource_tracker.py:279: UserWarning: resource_tracker: There appear to be 1 leaked semaphore objects to clean up at shutdown
  warnings.warn('resource_tracker: There appear to be %d '

Suggested Fix

Search for OpenMP associated with PyTorch.

$ find .venv -type f \( -name 'libomp.dylib' -o -name 'libgomp*.dylib' \) -print | grep torch
.venv/lib/python3.<PYVERSION>/site-packages/torch/lib/libomp.dylib

Create an .env

Create a .env file and add the OpenMP path as an environment variable, as shown below.

DYLD_LIBRARY_PATH=".venv/lib/python3.<PYVERSION>/site-packages/torch/lib"

Run uv

Pass the following --env-file options when running uv run.

uv run --env-file /path/to/.env ... python your_script.py