[Question] Running headless on Google Colab (Linux Build / CUDA 12.8)

[RaphalRiad]

Hi everyone,

I've been looking through the Wiki and I'm very interested in trying out LichtFeld Studio. Since I don't have a local Windows machine with an RTX GPU, I am hoping to run the software headless on Google Colab (using the standard T4 or A100 instances).

I reviewed the Installation and Command Line Options pages. Since binaries are currently Windows-only, I understand I need to build from source on Linux.

My Question:
Has anyone successfully set up a Colab notebook or a Linux installation script for this yet?

The Challenge:
The requirements listed (GCC 14+, CUDA 12.8, CMake 3.30+) are quite a bit newer than the default Google Colab environment (usually Ubuntu 22.04 with GCC 11/12). While I can try to upgrade the compilers manually via apt, I wanted to check if anyone has already solved the dependency chain (especially LibTorch and vcpkg) for a cloud environment.

Goal:
To run the CLI training headless, similar to:
./build/LichtFeld-Studio -d /content/data -o /content/output --strategy mcmc

If anyone has a .ipynb example or a snippet of the setup commands, it would be incredibly helpful for those of us without local high-end hardware!

Thanks for the hard work on the project!

[miriameng]

Isn't Google Colab meant for python instead of C++?