In the current version of cupy-xarray, there is a dependency on cupy.
SInce cupy 14, we can now install cupy as a wheel that will bring a ctk as "cupy-cuda12x[ctk]" which is very helpful for envs taht do not have a system cuda-toolkit.
However, when I tried to install via uv pip "cupy-cuda12x[ctk]" and then cupy-xarray I get this error that you can reproduce with a pyroject.toml like
[project]
name = "xarray-uv-example"
version = "0.1.0"
requires-python = "==3.13.*"
dependencies = [
"xarray",
"netCDF4",
"bottleneck",
"numpy",
"cupy-cuda12x[ctk]",
"cupy-xarray",
"jupyterlab-nvdashboard",
"ipykernel",
] and doing uv sync
Using CPython 3.13.12
Removed virtual environment at: .venv
Creating virtual environment at: .venv
Resolved 117 packages in 715ms
× Failed to build `cupy==14.0.1`
├─▶ The build backend returned an error
╰─▶ Call to `setuptools.build_meta.build_wheel` failed (exit status: 1)
[stdout]
Generating CUPY_CACHE_KEY from header files...
CUPY_CACHE_KEY (2006 files matching
/ephemeral/cache/uv/sdists-v9/pypi/cupy/14.0.1/Vn6dXZBXUzYnKgja7Fe3-/src/cupy/_core/include/**):
bb0abd009e32cb3266dc9bbd435ee72bc24f9575
-------- Configuring Module: cuda --------
command '/usr/bin/c++' failed with exit code 1
**************************************************
*** WARNING: Cannot check compute capability
command '/usr/bin/c++' failed with exit code 1
**************************************************
************************************************************
* CuPy Configuration Summary *
************************************************************
Build Environment:
Include directories: ['/ephemeral/cache/uv/sdists-v9/pypi/cupy/14.0.1/Vn6dXZBXUzYnKgja7Fe3-/src/cupy/_c
ore/include/cupy/_cccl/libcudacxx',
'/ephemeral/cache/uv/sdists-v9/pypi/cupy/14.0.1/Vn6dXZBXUzYnKgja7Fe3-/src/cupy/_core/include/cupy/_cccl/t
hrust',
'/ephemeral/cache/uv/sdists-v9/pypi/cupy/14.0.1/Vn6dXZBXUzYnKgja7Fe3-/src/cupy/_core/include/cupy/_cccl/cub',
'/ephemeral/cache/uv/sdists-v9/pypi/cupy/14.0.1/Vn6dXZBXUzYnKgja7Fe3-/src/cupy/_core/include',
'/home/ubuntu/.cache/uv/builds-v0/.tmp1RF3D8/lib/python3.13/site-packages/numpy/_core/include']
Library directories: []
nvcc command : (not found)
hipcc command : (not found)
Environment Variables:
CFLAGS : (none)
LDFLAGS : (none)
LIBRARY_PATH : (none)
CUDA_PATH : (none)
NVCC : (none)
HIPCC : (none)
ROCM_HOME : (none)
Modules:
cuda : No
-> Include files not found: ['cublas_v2.h', 'cuda.h', 'cuda_profiler_api.h', 'cuda_runtime.h',
'cufft.h', 'curand.h', 'cusparse.h']
-> Check your CFLAGS environment variable.
ERROR: CUDA could not be found on your system.
HINT: You are trying to build CuPy from source, which is NOT recommended for general use.
Please consider using binary packages instead.
Please refer to the Installation Guide for details:
https://docs.cupy.dev/en/stable/install.html
************************************************************
[stderr]
/tmp/tmpv5_2378z/a.cpp:1:10: fatal error: cublas_v2.h: No such file or directory
1 | #include <cublas_v2.h>
| ^~~~~~~~~~~~~
compilation terminated.
/tmp/tmph5zfrr4x/a.cpp:2:18: fatal error: cuda_runtime_api.h: No such file or directory
2 | #include <cuda_runtime_api.h>
| ^~~~~~~~~~~~~~~~~~~~
compilation terminated.
Traceback (most recent call last):
File "<string>", line 14, in <module>
requires = get_requires_for_build({})
File
"/home/ubuntu/.cache/uv/builds-v0/.tmp1RF3D8/lib/python3.13/site-packages/setuptools/build_meta.py", line 333,
in get_requires_for_build_wheel
return self._get_build_requires(config_settings, requirements=[])
~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File
"/home/ubuntu/.cache/uv/builds-v0/.tmp1RF3D8/lib/python3.13/site-packages/setuptools/build_meta.py", line 301,
in _get_build_requires
self.run_setup()
~~~~~~~~~~~~~~^^
File
"/home/ubuntu/.cache/uv/builds-v0/.tmp1RF3D8/lib/python3.13/site-packages/setuptools/build_meta.py", line 317,
in run_setup
exec(code, locals())
~~~~^^^^^^^^^^^^^^^^
File "<string>", line 56, in <module>
File "/ephemeral/cache/uv/sdists-v9/pypi/cupy/14.0.1/Vn6dXZBXUzYnKgja7Fe3-/src/install/cupy_builder/cup
y_setup_build.py", line 560, in get_ext_modules
extensions = make_extensions(ctx, compiler, use_cython)
File "/ephemeral/cache/uv/sdists-v9/pypi/cupy/14.0.1/Vn6dXZBXUzYnKgja7Fe3-/src/install/cupy_builder/cup
y_setup_build.py", line 412, in make_extensions
raise Exception('Your CUDA environment is invalid. '
'Please check above error log.')
Exception: Your CUDA environment is invalid. Please check above error log.
hint: This usually indicates a problem with the package or the build environment.
help: `cupy` (v14.0.1) was included because `xarray-uv-example` (v0.1.0) depends on `cupy-xarray` (v0.1.4)
which depends on `cupy` I think it's because uv has no way of knowing that cupy-cuda12x already satisfies the cupy import.
I noticed that currently on main, there is no cupy dependency, this was removed on this commit 3f84de0
Which is months after the later release. I tried installing from main, and this seem to work. So this would probably solve with a release cut.
NOTE:
- I wonder if cupy should be made a dependency again, given that you can't use it without cupy.
- Not sure how, but if there is a way to make sure that when installed with pip it depends on "cupy-cuda12x[ctk]" or
"cupy-cuda13x[ctk]" depending on the cuda version. This will bring a ctk, similarly on how it happens in conda
cc: @jacobtomlinson @jayavenkatesh19
In the current version of cupy-xarray, there is a dependency on
cupy.SInce cupy 14, we can now install cupy as a wheel that will bring a ctk as
"cupy-cuda12x[ctk]"which is very helpful for envs taht do not have a system cuda-toolkit.However, when I tried to install via uv pip "cupy-cuda12x[ctk]" and then
cupy-xarrayI get this error that you can reproduce with apyroject.tomllikeand doing
uv syncI think it's because uv has no way of knowing that cupy-cuda12x already satisfies the cupy import.
I noticed that currently on main, there is no
cupydependency, this was removed on this commit 3f84de0Which is months after the later release. I tried installing from main, and this seem to work. So this would probably solve with a release cut.
NOTE:
"cupy-cuda13x[ctk]"depending on the cuda version. This will bring a ctk, similarly on how it happens in condacc: @jacobtomlinson @jayavenkatesh19