Build Error

[zang09]

Hi, thanks for sharing amazing works!

Even after following the instructions in the README.md and building the Dockerfile exactly as provided, I still encounter the following error at runtime:

RuntimeError:
The detected CUDA version (11.8) mismatches the version that was used to compile
PyTorch (12.6). Please make sure to use the same CUDA versions.

I’m working inside a Conda environment, and torch.version.cuda confirms that my PyTorch is built with CUDA 11.8. I also tried purging the cache (~/.cache/torch_extensions) and rebuilding, but the issue persists.

From what I can tell, the problem seems to originate from the recently updated gsplat dependency, specifically the integration with the new SplatAD components. This issue didn’t appear in earlier commits before the update.

Could you please take a look and confirm whether the current gsplat setup is compatible with CUDA 11.8 and PyTorch 2.0.1? Any guidance would be appreciated.

[sheelabhadra]

I ran into a similar issue. Found a workaround by adding the following line to Dockerfile after line 77

RUN sed -i 's|gsplat@git+https://github.com/carlinds/splatad.git|gsplat|' pyproject.toml

Hope it helps!