Install gsplat with manually specified/multiple CUDA architectures and multiple compute capabilities

[Zunhammer]

Hi,

I'm including gplsat in a docker image but it seems that it defines the CUDA architecture during docker build to my local machine where everything is running fine. However, I cannot use the image on other machines where I get the error:
'''gsplat: No CUDA toolkit found. gsplat will be disabled.'''

Is there a way to build gsplat for multiple CUDA architectures and compute capabilities? Usually, I can do this using environment variables or by passing the specific architectures to the build command (e.g. CUDA_ARCHITECTURES=90;89;86;80;75;70;61;52 for tiny-CUDA-nn).

Thank you for your support.