Troubleshooting

Known Issues

• The Triton-fused FusedVolumeRenderer requires CUDA and a working Triton installation. On Windows the triton-windows fork is used automatically — if Triton is unavailable, keep triton=False (the default) so rendering falls back to the pure-PyTorch VolumeRaycaster path.

• FusedVolumeRenderer explicitly disables CUDA autocasting; controlling the accumulation dtype manually. If you rely on mixed precision, do your up/down-casting around the render call.

Frequently Asked Questions

Go ask us some questions on GitHub!