An error occurs when running splatfacto in the nerfstduio environment of docker：RuntimeError: CUDA error: no kernel image is available for execution on the device CUDA kernel errors might be asynchronously reported at some other API call

[luanyu3]

No description provided.

[SmthFail]

@luanyu3 I have this error too. I run docker on different systems.
On NVIDIA RTX 2080 - work,
On NVIDIA A100 - work,
On NVIDIA RTX 4090 - didn't work,
On NVIDIA V100 - didn't work.

After one week of research and rebuilds docker only find this solution: use precompiled libs from gsplat.
To achieve this change 80 line in Dockerfile from this

pip install --no-cache-dir git+https://github.com/nerfstudio-project/gsplat.git@v${GSPLAT_VERSION} && \

to this

pip install --no-cache-dir ninja 'numpy<2.0.0' jaxtyping rich && \
pip install --no-cache-dir gsplat --index-url https://docs.gsplat.studio/whl/pt21cu118 && \

then rebuild docker image. After this my docker image work on all platform without problems. Also it decrease docker image build time