Spaces:
Running on Zero
Running on Zero
[Admin maintenance] Migrate grant to ZeroGPU
#35
by multimodalart HF Staff - opened
Bundled torchmcubes wheel was built against CUDA 11 (libcudart.so.11.0), incompatible with the current ZeroGPU image (CUDA 12). Build from source instead, with TORCH_CUDA_ARCH_LIST=12.0+PTX so kernels target Blackwell (sm_120). Adds scikit-build-core/pybind11/cmake/ninja so the source build can run.
multimodalart changed pull request status to merged