You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Description
When trying to use this project on Windows with an RTX 4080, I noticed that the documentation recommends torch 2.8.0. However, in practice, GSPLAT's CUDA extensions do not seem to compile successfully with PyTorch 2.8.0.
Issues encountered
Errors like ImportError: cannot import name 'csrc' from 'gsplat'
Runtime errors during CUDA extension compilation
_jit_compile() missing 2 required positional arguments: 'verbose' and 'with_sycl'
It would be helpful if the documentation or requirements could clarify:
The currently recommended PyTorch + CUDA combination
That using the latest PyTorch version may cause GSPLAT compilation to fail
Environment
Windows 11
Python 3.13
PyTorch 2.8.0+ CUDA 12.4
RTX 4080
Community examples and some testing indicate that GSPLAT extensions can successfully build with torch 2.1.2 + CUDA 11.8. This might be useful information for users trying to run this project on Windows. nerfstudio-project/gsplat#311
Am I out of date?
Request
I am looking for guidance or instructions on how to successfully install and compile GSPLAT under these conditions. Any help would be appreciated.