Open PierceLBrooks opened 1 year ago
I think the add-on should somehow break free from those dependencies, but as a quickfix I might add a torch version parameter in the preferences
At least in the case of support for CUDA 10.2, the only version of torch that offers a Python 3.10 wheel build is 1.12.0, and even then only for linux platforms: https://pytorch-geometric.com/whl/torch-1.12.0+cu102.html
FYI I also tried being sly about it and editing the find link flag for the virtual environment's PIP package installation invokations ( https://github.com/pKrime/brignet/blob/v0.1-alpha/setup_utils/venv_utils.py#L260 ) to an edited local copy of the wheel build directory page that pointed to an appropriate local wheel build, which somehow still did not seem to work sadly.
I had some issues compiling sparse torch on Windows for python 3.10
Something in the C++ extension compiling is failing.
C:\Users\ernes\micromamba\envs\brignet\lib\site-packages\torch\include\pybind11\cast.h(624): error: too few arguments for template template parameter "Tuple"
detected during instantiation of class "pybind11::detail::tuple_caster<Tuple, Ts...> [with Tuple=std::pair, Ts=<T1, T2>]"
(721): here
C:\Users\ernes\micromamba\envs\brignet\lib\site-packages\torch\include\pybind11\cast.h(717): error: too few arguments for template template parameter "Tuple"
detected during instantiation of class "pybind11::detail::tuple_caster<Tuple, Ts...> [with Tuple=std::pair, Ts=<T1, T2>]"
(721): here
The rest of the dependencies seems to go through for brignet
This still remains an issue for Blender 4.0+ builds.
I've been using https://github.com/V-Sekai/blender-rignet as my fork with Blender 4 if this helps anyone.
The version 3.1 release patch notes for Blender show the internal distribution of Python has been bumped up to 3.10: https://wiki.blender.org/wiki/Reference/Release_Notes/3.1/Python_API#Python_3.10
This causes problems for resolving dependencies upon installation through the pip package manager using the provided wheel module listing, which only offers support for the necessary CUDA versions up to Python 3.9: https://download.pytorch.org/whl/torch_stable.html
Here is some terminal output from an installation attempt through the Blender UI to illustrate this: