Ubuntu 23.10 successful install 😄

[watsieboi]

I successfully installed on Ubuntu without deviating from the original install instructions too much. Here's how I got this code working:

• conda create --name comfy -y python=3.11
• conda activate comfy
• python -m pip install --upgrade pip
• conda install -c "nvidia/label/cuda-12.1.0" cuda-toolkit (I've found this to be more successful than globally installing CUDA when you need different versions for different environments)

From there I followed the normal install instructions and "Install Plan B". That got me 95% of the way there, then it was a matter of installing a couple random dependencies as they popped up (matplotlib and some others). Next I downloaded the zero123 model and boom the code runs beautifully. Thank you MrForExample for the amazing implementation, this is truly groundbreaking!! I'm inspired to experiment with building my own GS related nodes in ComfyUI :)

[MrForExample]

Thanks for your effort my friend, I'll add that into install instructions soon, cheers :)
Let me know if there is anything I maybe able to help you with your own GS related nodes

[vicentecarro]

How to do it with venv + pip? I think this should be the default way, right?

[MrForExample]

How to do it with venv + pip? I think this should be the default way, right?

The miniconda is basically venv + build environment manager + packages manager, there are many packages that need proper setup in order to build successfully, diff-gaussian-rasterization, nvdiffrast, pytorch3d just to name a few. That's why it's a recommend way of install Comfy3D and most of other Machine Learning projects really.

[OrvilleQ]

Just a reminder here.

If you are using a cutting-edge distribution like Arch or OpenSUSE Tumbleweed, you might have a GCC version that is incompatible with the current NVCC versions. This incompatibility can cause issues when installing post requirements. To fix this issue, you need to install an older version of GCC that is supported by your NVCC using Conda.

conda install gcc_linux-64=<the version you need> gxx_linux-64=<the version you need> -c conda-forge

You can check the CUDA documentation to find the GCC version you need.