Skip to content

Ecosystem compatibility with numpy 2.0 #26191

Closed

Description

@rgommers

This list tracks the compatibility status of packages that depend on or support NumPy. If "compatible release on PyPI" does not say "yes" but a version number is listed: this is based on plans announced in a tracking issue or other communication by the authors of the package.

Maintainers: please feel free to edit directly (please refresh the page first to avoid overwriting edits from others!). Others who want to update things: please do comment, or feel free to ping me elsewhere.

Package name Compatible release on PyPI? Min compatible version Notes
Adaptive yes 1.3.0 python-adaptive/adaptive#458
arch yes 7.0.0 bashtage/arch#720
AstroPy yes 6.1.0 astropy/astropy#16200
astropy-healpix yes 1.0.3 astropy/astropy-healpix#214
autograd yes 1.7.0 HIPS/autograd#618
AwkwardArray yes 2.6.3 scikit-hep/awkward#3064
basemap yes 2.0.0 matplotlib/basemap#604
BioPython yes 1.84 biopython/biopython#4676
Bokeh yes 3.4.1 bokeh/bokeh#13835
boost-histogram yes 1.4.1 #26191 (comment)
Boost.Python yes 1.87.0 boostorg/python#431
bottleneck yes 1.4.0 pydata/bottleneck#453
CARMA n/a 0.8.0 RUrlus/carma#129
Cartopy yes 0.23 SciTools/cartopy#2339
Catboost yes 1.2.8 catboost/catboost#2671
cftime yes 1.6.4.post1 python-adaptive/adaptive#458
clawpack yes 5.11.0 clawpack/clawpack#252
cmocean yes 3.1.0 matplotlib/cmocean#99
ContourPy yes 1.2.1 contourpy/contourpy#371
CPNest yes 0.11.6 johnveitch/cpnest#92
CuPy yes 14.0.0 (>=13.2 partial support) cupy/cupy#8306
Cython yes 3.0.4 Version is an estimate, it's worked fine for quite a while
Cython-BLIS yes 1.0.0 explosion/cython-blis#106
Dask yes 2024.5.1 dask/dask#11066
Datashader yes 0.16.2 holoviz/datashader#1324
dpnp yes 0.16.0 IntelPython/dpnp#1861
EigenPy yes 3.10.0 stack-of-tasks/eigenpy#496
ewah_bool_utils yes 1.2.2 NumPy v2 support first introduced in https://github.com/yt-project/ewah_bool_utils/releases/tag/v1.2.0, and then, better NumPy v2 Windows support introduced in yt-project/ewah_bool_utils#101
fastparquet yes 2024.5.0 dask/fastparquet#922
fitsio yes 1.2.3 esheldon/fitsio#393
GDAL yes 3.9.0 OSGeo/gdal#9751
genai-perf yes 0.0.11 triton-inference-server/perf_analyzer#275
GeoPandas yes 0.14.4 geopandas/geopandas#3258
GNU Radio n/a 3.10.10.1 gnuradio/gnuradio#7378
h5py yes 3.11.0 h5py/h5py#2353
holoviews yes 1.19.0 holoviz/holoviews#6238
hypothesis yes 6.100.2 HypothesisWorks/hypothesis#3950
imagecodecs yes 2024.6.1 cgohlke/imagecodecs#100
Imageio yes 2.34.2 imageio/imageio#1090
iminuit yes 2.26.0 scikit-hep/iminuit#977
ITK yes 5.4.0 InsightSoftwareConsortium/ITK#4700
JAX yes 0.4.26 jax-ml/jax#19246
jupyterlite-pyodide-kernel yes 0.4.7 jupyterlite/pyodide-kernel#156
Keras yes 3.5.0 keras-team/keras#19691
LightGBM yes 4.4.0 lightgbm-org/LightGBM#6454 & lightgbm-org/LightGBM#6439
lightly yes 1.5.11 lightly-ai/lightly#1558
linearmodels yes 6.0.0 bashtage/linearmodels#593
Matplotlib yes 3.8.4 matplotlib/matplotlib#26778
MDAnalysis yes 2.8.0 MDAnalysis/mdanalysis#4482
ml_dtypes yes 0.4.0 jax-ml/ml_dtypes#143
mlpack yes 4.5.0 mlpack/mlpack#3723
MNE-Python yes 1.7.0 mne-tools/mne-python#12672
Modin yes 0.31.0 modin-project/modin#7310
msprime yes 1.3.2 tskit-dev/msprime#2287
netCDF4 yes 1.7.0 Unidata/netcdf4-python#1317
NetworkX yes 3.3 networkx/networkx#7390
ngboost yes 0.5.2 stanfordmlgroup/ngboost#358
Nipy yes 0.6.1 nipy/nipy#565
Numba yes 0.60 numba/numba#9544, Discourse post with context
numcodecs yes 0.12.1 Likely older versions too; has been stable for a while. zarr-developers/numcodecs#521
numexpr yes 2.10.0 pydata/numexpr#478
numpngw yes 0.1.4
OpenCV yes 4.10.0.84 opencv/opencv-python#943
Outlines yes 0.1.6 dottxt-ai/outlines#1104
Pandas yes 2.2.2 pandas-dev/pandas#55519
pint yes 0.24 hgrecco/pint#1974
Polars yes 1.1.0 pola-rs/polars#16998
polyagamma yes 2.0.0 zoj613/polyagamma#127
pvlib yes 0.10.5 pvlib/pvlib-python#2026
PyArrow yes 16.0 apache/arrow#39532
Py-ART / arm-pyart yes 1.18.2 ARM-DOE/pyart#1550
Pybind11 yes 2.12.0 pybind/pybind11#5009
PyData Sparse yes 0.15.2 pydata/sparse#680
pyEDFlib yes 0.1.38 holgern/pyedflib#259
PyGSL yes 2.5.1
PyKrige yes 1.7.2 GeoStat-Framework/PyKrige#290
PyMC yes 5.21.0 pymc-devs/pymc#7688
Pyodide yes 0.27.0 pyodide/pyodide#4925
Pyomo yes 6.8.0 Pyomo/pyomo#3292
PySide6 yes 6.7.0 from a comment on this issue: #26191 (comment)
PySpark yes 4.0.0 apache/spark#47083
PyTables yes 3.10 PyTables/PyTables#1083
PyTensor yes 2.28.0 pymc-devs/pytensor#689
Pythran yes 0.16.0 (0.15.0 works mostly, SciPy builds with it) serge-sans-paille/pythran#2189
PyTorch yes 2.3.0 (*) pytorch/pytorch#107302 (Windows binaries issue with 2.3.x-2.4.0: pytorch#131668
PyVista yes 0.44.0 pyvista/pyvista/releases/tag/v0.44.0
PyWavelets yes 1.6.0 PyWavelets/pywt#731
PyXIRR yes 0.10.5 https://github.com/Anexen/pyxirr/releases/tag/v0.10.5 via Anexen/pyxirr@9353db2
QuTiP yes 5.0.3 qutip/qutip#2421, qutip/qutip#2457
randomgen yes 2.0.0 bashtage/randomgen#375
RAPIDS yes 24.10 rapidsai/build-planning#38
Rasterio yes 1.3.10 rasterio/rasterio#3024
RDKit yes 2024.3.4 rdkit/rdkit#7477
reikna yes  0.9.0  fjarri/reikna#69
river yes 0.22.0 online-ml/river#1632
rust-numpy yes 0.22 PyO3/rust-numpy#409
sagemaker yes 2.254.0 aws/sagemaker-python-sdk#4882
scikit-bio yes 0.6.1 scikit-bio/scikit-bio#1964
scikit-image yes 0.23.1 scikit-image/scikit-image#7282
scikit-learn yes 1.4.2 scikit-learn/scikit-learn#27075
scikit-sparse yes 0.4.16 scikit-sparse/scikit-sparse#120
scipp yes 24.06.0 scipp/scipp#3427
SciPy yes 1.13.0 scipy/scipy#20375
Shapely yes 2.0.4 shapely/shapely#1972
SpaCy yes 3.8.2 explosion/spaCy#13528
spglib yes 2.5.0 spglib/spglib#407
SymPy yes 1.12.1
Seaborn yes 0.13.2 mwaskom/seaborn#3683
statsmodels yes 0.14.2 statsmodels/statsmodels#9194
TensorBoard yes 2.18.0 tensorflow/tensorboard#6869
TensorFlow yes 1.18.0 tensorflow/tensorflow#67291, has <2 upper bound for 2.16.1, requirements, lock file
Thinc yes 9.1.1 explosion/thinc#939
threadpoolctl yes 3.5.0 joblib/threadpoolctl#175
tifffile yes 2024.4.24 cgohlke/tifffile#252
TorchGeo yes 0.6.0 torchgeo/torchgeo#2151
torchvision yes 0.19.1 pytorch/vision#8460
TotalSegmentor yes 2.11.0 wasserth/TotalSegmentator#504
treelite yes 4.2.1 dmlc/treelite#560
tritonclient yes 2.56.0 triton-inference-server/client#792
unyt yes 3.0.2 yt-project/unyt#493
VisPy yes 0.14.3 vispy/vispy#2599
vLLM yes 0.8.3 vllm-project/vllm#6570
wavio yes 0.0.9
Xarray yes 2024.06.0 pydata/xarray#8844
XGBoost yes 2.1.0 dmlc/xgboost#10221
yt yes 4.3.1 yt-project/yt#4859
Zarr yes 2.18.0 zarr-developers/zarr-python#1818
zfpy yes 1.0.1 (PyPI, not the 1.0.1 tag on the repo) llnl/zfp#210

Activity

  1. pinned this issue on Apr 1, 2024
  2. added this to the 2.0.0 release milestone on Apr 1, 2024
  3. jakevdp commented on Apr 1, 2024

    @jakevdp
    Contributor

    Thanks - for JAX we are planning an 0.4.26 release in the next day or two which will be built against NumPy 2.0.0rc1.

  4. jakevdp commented on Apr 1, 2024

    @jakevdp
    Contributor

    Also, in case you want to add it, we just released ml_dtypes v0.4.0 (https://pypi.org/project/ml-dtypes/) which is compatible with NumPy 2.0.

  5. jakirkham commented on Apr 1, 2024

    @jakirkham
    Contributor

    Thanks for putting this together Ralf! 🙏

    This is incredibly helpful 🙂

  6. rgommers commented on Apr 1, 2024

    @rgommers
    MemberAuthor

    Thanks @jakevdp, I added the info for both JAX and ml_dtypes.

  7. neutrinoceros commented on Apr 2, 2024

    @neutrinoceros
    Contributor

    unyt 3.0.2 was also released a couple days ago for compat with numpy 2, in case you'd like to include it !

  8. neutrinoceros commented on Apr 2, 2024

    @neutrinoceros
    Contributor

    Also, here's where to track progress for cartopy : SciTools/cartopy#2339

  9. rgommers commented on Apr 2, 2024

    @rgommers
    MemberAuthor

    Thanks @neutrinoceros, much appreciated. Added both packages.

  10. ianthomas23 commented on Apr 2, 2024

    @ianthomas23
    Contributor

    ContourPy 1.2.1 has just been released on PyPI (https://pypi.org/project/contourpy/1.2.1/) with NumPy 2 compatibility, most relevant PR is contourpy/contourpy#371. It is a compulsory dependency of Matplotlib.

  11. jorisvandenbossche commented on Apr 3, 2024

    @jorisvandenbossche
    Contributor

    For PyArrow, it will be the upcoming 16.0 release that will be the first numpy-2.0-compatible release, but this will only be expected in around 3 weeks at the earliest (but normally certainly before the end of the month).
    For people that need a PyArrow installed in an environment with numpy 2.0, it's worth noting that there are nightly wheels available though that already work with 2.0.

  12. dkbarn commented on Apr 3, 2024

    @dkbarn

    Could we add OpenCV to this list? I filed a ticket here to track progress on a numpy 2.0 build for it.

  13. rgommers commented on Apr 3, 2024

    @rgommers
    MemberAuthor

    Thanks @ianthomas23, @jorisvandenbossche, @dkbarn - all info added to the table.

  14. hawkinsp commented on Apr 3, 2024

    @hawkinsp
    Contributor

    JAX released 0.4.26 on pypi, which is compatible with NumPy 2.0.

    (NumPy folks: congratulations on your imminent v2 release!)

  15. 153 remaining items

  16. dyollb commented on Aug 25, 2025

    @dyollb
  17. rgommers commented on Sep 26, 2025

    @rgommers
    MemberAuthor

    Okay, it's time. I will close this issue and declare victory on the ecosystem transitioning to NumPy 2.0. 🎉

    The updates on this issue have slowed down a lot over the past couple of months, and of the many packages we are tracking in this issue, only one didn't complete its support yet.

  18. jakirkham commented on Oct 30, 2025

    @jakirkham
    Contributor

    This one is now complete

    Package name Compatible release on PyPI? Min compatible version Notes
    sagemaker Y 2.254.0 aws/sagemaker-python-sdk#5311
  19. rgommers commented on Oct 30, 2025

    @rgommers
    MemberAuthor

    Thanks @jakirkham. I updated the table with that, given that it was the very last existing entry that still didn't have support until now. So now we're really really done:)

  20. mhvk commented on Dec 4, 2025

    @mhvk
    Contributor

    Sorry for what may be highjacking this thread, but according to SPEC 0, we're at the end for numpy 1.x support now. Astropy right now is planning to drop support in April, but continuing to support our current 7.2 for an extra half year, to ease the transition.

    I personally would really like to drop 1.x for regular development, since numpy 2.x allows some really nice performance improvements, but there is some worry we are jumping the gun, and no-one else is planning to drop 1.x in the next half year. Is this the case? What are people planning to do (especially interested in scipy, which is one of our most used optional dependencies).

    Note that there was some discussion in scientific-python/specs#386 about making an exception to SPEC 0, but that died out. Fine to be referred to that!

  21. andyfaff commented on Dec 4, 2025

    @andyfaff
    Member

    What are people planning to do (especially interested in scipy, which is one of our most used optional dependencies)

    It's not clear what you mean by this? Scipy still has compat for 1.26.4, but I don't think there's a problem with binary compatibility.
    I wouldn't see any drawback with making a minimum 2.0 decision for astropy. It's a question of what your downstream project would want.

  22. jakevdp commented on Dec 4, 2025

    @jakevdp
    Contributor

    What are people planning to do (especially interested in scipy, which is one of our most used optional dependencies).

    In JAX, we follow SPEC0 for numpy/scipy dependencies. We dropped NumPy 1.X support for the v0.7.2 release in September, which required bumping the minimum SciPy version to 1.13 four months early. It's been almost 3 months since this release, and we haven't heard any complaints that I'm aware of.

  23. mhvk commented on Dec 4, 2025

    @mhvk
    Contributor

    @andyfaff - I meant whether for a release in the next half year, you are planning to set a minimum numpy requirement of >=2.0. In astropy, the PR removing 1.x support removes about 1000 lines (astropy/astropy#18986), but for me the main benefit is being able to use new features (e.g., user dtypes).

    And, yes, you are right, it is a question about what downstream would want, but this is not that easy to answer! My sense is that if one is stuck on numpy 1.26, it doesn't hurt to be stuck on astropy 7.2 as well, but others disagree... Anyway, really just trying to get a sense of whether people are more or less following SPEC 0 or not.

    @jakevdp - thanks! Really good to hear you didn't have major problems.

  24. andyfaff commented on Dec 4, 2025

    @andyfaff
    Member

    Currently the next scipy release (1.17 before Christmas) will have a minimum runtime of numpy>=1.26.4, but the build system requires numpy>=2.0.0. At least that's what the pyproject.toml says, I can't remember a discussion on making 2.0 the minimum runtime. In CI we test across main >= numpy.version.version >= 1.26.4.
    From SPEC0 it does look like we should drop 1.26.4 for 1.18.

  25. mhvk commented on Dec 4, 2025

    @mhvk
    Contributor

    Yes, some of the people looked through larger projects and noticed they had not upped the runtime requirement to 2.0 yet, so we were curious whether people are intending to follow SPEC 0, or explicitly wait a little longer.

  26. lesteve commented on Dec 5, 2025

    @lesteve
    Contributor

    As a data point for the scikit-learn more conservative policy (at least the current consensus see our maintainer doc), it seems like we will be dropping numpy 1.x around November 2027 (i.e. in roughly 2 years). To be explicit, I don't think we are planning to have a numpy 2 specific rule, but we haven't really discussed it amongst scikit-learn maintainers ...

    I don't think this is an issue per se that scikit-learn is more conservative, but if somehow you see undesired side-effects for the wider ecosystem, let us know. At the same time, to manage expectations, I won't have time to engage in long discussions about this topic 1 until beginning January.

    Explanation (see our current rule):

    • numpy 1.26 supports Python 3.12
    • We will be requiring Python>=3.13 in November 2027. Python 3.13 was released on 2024-10-07 and our minimum Python is the one that is a bit more than 3 years old (Python has a yearly release around October)
    • around November 2027, we will require Python>=3.13, numpy>=2.1.0

    Footnotes

    1. for some reason I spent some effort to have scikit-learn write its own explicit rule in the context of a SPEC0 opt-in LTS variant (seems like there is some kind of movement recently, not sure what exactly ... https://github.com/scientific-python/specs/pull/389). To be honest, I don't find this topic fun or interesting so at this point I am kind of hoping to stay away from long-winded discussions around it 😅. ↩

  27. mattip commented on Dec 5, 2025

    @mattip
    Member

    I think the discussion about dropping NumPy 1 is off-topic for this isssue, which is closed. If needed, please open a new issue or even better discuss it on one of the scientific-python forums or issue trackers.

  28. rgommers commented on Dec 6, 2025

    @rgommers
    MemberAuthor

    @mhvk I added a comment at scientific-python/specs#386 (comment). It's too early for dropping 1.26 for core (widely used, general-purpose) projects. AstroPy has a slightly more niche audience, so no strong opinion from me there.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions