Someone treated already a similar issue ?

Hey. How did you install the package (pip, conda)? Also, can you provide the code that you're running? We test multithreading for macos arm64 here. The cases where I've seen this segfault is when you have scikit-learn/numpy/scipy from conda and install mlforecast from pip

Sorry for my late reply. Every single package and dependencies managed by pip on isolated pyenv environment

Hmm, it's hard to diagnose. It's most likely due to conflicting openmp versions being loaded. If you don't have many series it won't make much of a difference, so setting num_threads=1 should be fine. If you do want to use multithreading you can try installing from github (

pip install git+<https://github.com/nixtla/coreforecast|https://github.com/nixtla/coreforecast>

), which will compile the library locally and link the correct openmp lib, note that this requires a c++ compiler and ninja.

Awesome !!! i will test the solution today and i will reply back. Thanks for your help

Thanks for the updates. So the fix was to install openmp?

Exactly and when provide the correct path to compiler adding recommended flag in the shell user profile file

I see. We use clang to compile for macos and link brew's libomp, however we also include libomp in the wheels which may be what is causing the conflict. Sorry for the troubles. So it's working for you now?

Yes it’s working fine. I will test the solution with clang compiler on one other station, I will write you back