Problem with GPU (MPS)

[paularandamar]

Before you fill out this form:

Did you review the FAQ?

• [x ] Yes, I reviewed the FAQ on the cellpose ReadTheDocs

Did you look through previous (open AND closed) issues posted on GH?

• [x ] Yes, I searched for related previous GH issues.

Now fill this form out completely:

Install problem

Cellpose installs correctly, but when launching the GUI it fails during model loading.

The main error is:

ERROR: BFloat16 is not supported on MPS

This happens on macOS when Cellpose automatically uses the MPS GPU backend.

In addition, there are NumPy compatibility warnings at startup due to a mismatch between NumPy 2.x and compiled dependencies.

Environment info

Details

Please run `conda list` in your cellpose environment in the terminal / anaconda prompt to let us know your software versions.

bzip2                      1.0.8            h6c40b1e_6
ca-certificates            2025.7.15        hecd8cb5_0
cellpose                   4.1.1            pypi_0           pypi
colorama                   0.4.6            pypi_0           pypi
expat                      2.7.1            h6d0c2b6_0
fastremap                  1.19.0           pypi_0           pypi
filelock                   3.29.2           pypi_0           pypi
fill-voids                 2.1.2            pypi_0           pypi
fsspec                     2026.4.0         pypi_0           pypi
imagecodecs                2026.6.6         pypi_0           pypi
jinja2                     3.1.6            pypi_0           pypi
libcxx                     19.1.7           haebbb44_3
libffi                     3.4.4            hecd8cb5_1
markupsafe                 3.0.3            pypi_0           pypi
mpmath                     1.3.0            pypi_0           pypi
natsort                    8.4.0            pypi_0           pypi
ncurses                    6.5              h923df54_0
networkx                   3.6.1            pypi_0           pypi
numpy                      2.4.6            pypi_0           pypi
opencv-python-headless     4.13.0.92        pypi_0           pypi
openssl                    3.0.17           hee2dfae_0
packaging                  26.2             pypi_0           pypi
pillow                     12.2.0           pypi_0           pypi
pip                        26.1.1           pyhc872135_1
pygments                   2.20.0           pypi_0           pypi
pyqt6                      6.11.0           pypi_0           pypi
pyqt6-qt6                  6.11.1           pypi_0           pypi
pyqt6-sip                  13.11.1          pypi_0           pypi
pyqtgraph                  0.14.0           pypi_0           pypi
python                     3.12.11          he8d2d4c_0
qtpy                       2.4.3            pypi_0           pypi
readline                   8.3              h49f2429_0
roifile                    2026.2.10        pypi_0           pypi
scipy                      1.17.1           pypi_0           pypi
segment-anything           1.0              pypi_0           pypi
setuptools                 78.1.1           py312hecd8cb5_0
sqlite                     3.50.2           hc8b0dd6_1
superqt                    0.8.2            pypi_0           pypi
sympy                      1.14.0           pypi_0           pypi
tifffile                   2026.6.1         pypi_0           pypi
tk                         8.6.15           h3a5a201_0
torch                      2.2.2            pypi_0           pypi
torchvision                0.17.2           pypi_0           pypi
tqdm                       4.68.2           pypi_0           pypi
typing-extensions          4.15.0           pypi_0           pypi
tzdata                     2026b            he532380_0
wheel                      0.45.1           py312hecd8cb5_0
xz                         5.6.4            h46256e1_1
zlib                       1.2.13           h4b97444_1

Run log

Please post all command line/notebook output for us to understand the problem. For this please make sure you are running with verbose mode on. So command line, with --verbose tag, or in a notebook first run

Details

A module that was compiled using NumPy 1.x cannot be run in NumPy 2.4.6 as it may crash. To support both 1.x and 2.x versions of NumPy, modules must be compiled with NumPy 2.0. Some module may need to rebuild instead e.g. with 'pybind11>=2.12'. If you are a user of the module, the easiest solution will be to downgrade to 'numpy<2' or try to upgrade the affected module. We expect that some modules will need time to support NumPy 2. 
Traceback (most recent call last): File "/Users/paulaarandamartinez/miniforge3/envs/cellpose/bin/cellpose", line 3, in <module> from cellpose.__main__ import main File "/Users/paulaarandamartinez/miniforge3/envs/cellpose/lib/python3.12/site-packages/cellpose/__init__.py", line 1, in <module> from cellpose.version import version, version_str File "/Users/paulaarandamartinez/miniforge3/envs/cellpose/lib/python3.12/site-packages/cellpose/version.py", line 7, in <module> import torch File "/Users/paulaarandamartinez/miniforge3/envs/cellpose/lib/python3.12/site-packages/torch/__init__.py", line 1477, in <module> from .functional import * # noqa: F403 File "/Users/paulaarandamartinez/miniforge3/envs/cellpose/lib/python3.12/site-packages/torch/functional.py", line 9, in <module> import torch.nn.functional as F File "/Users/paulaarandamartinez/miniforge3/envs/cellpose/lib/python3.12/site-packages/torch/nn/__init__.py", line 1, in <module> from .modules import * # noqa: F403 File "/Users/paulaarandamartinez/miniforge3/envs/cellpose/lib/python3.12/site-packages/torch/nn/modules/__init__.py", line 35, in <module> from .transformer import TransformerEncoder, TransformerDecoder, \ File "/Users/paulaarandamartinez/miniforge3/envs/cellpose/lib/python3.12/site-packages/torch/nn/modules/transformer.py", line 20, in <module> device: torch.device = torch.device(torch._C._get_default_device()), # torch.device('cpu'), /Users/paulaarandamartinez/miniforge3/envs/cellpose/lib/python3.12/site-packages/torch/nn/modules/transformer.py:20: UserWarning: Failed to initialize NumPy: _ARRAY_API not found (Triggered internally at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/utils/tensor_numpy.cpp:84.) device: torch.device = torch.device(torch._C._get_default_device()), # torch.device('cpu'), 2026-06-10 17:14:09,475 
[INFO] WRITING LOG OUTPUT TO /Users/paulaarandamartinez/.cellpose/run.log 2026-06-10 17:14:09,476 [INFO] cellpose version: 4.1.1 platform: darwin python version: 3.12.11 torch version: 2.2.2 2026-06-10 17:14:10,358 
[INFO] ** TORCH MPS version installed and working. ** 
GUI_INFO: loading image: /Volumes/Disco_Paula/Postdoc/Blaauw Lab/Microscopio/PAS/nerve crush 30d/30d/PAS/cellpose_control leg 2791_ch00.tif 
GUI_INFO: image shape: (8000, 9000, 3) 
GUI_INFO: normalization checked: computing saturation levels (and optionally filtered image) {'lowhigh': None, 'percentile': [1.0, 99.0], 'normalize': True, 'norm3D': True, 'sharpen_radius': 0.0, 'smooth_radius': 0.0, 'tile_norm_blocksize': 0.0, 'tile_norm_smooth3D': 0.0, 'invert': False} [np.float32(133.0), np.float32(245.0)] 2026-06-10 17:15:27,553 [INFO] ** TORCH MPS version installed and working. ** 2026-06-10 17:15:27,559 
[INFO] >>>> using GPU (MPS) 2026-06-10 17:15:29,298 [INFO] >>>> loading model /Users/paulaarandamartinez/.cellpose/models/cpsam 
ERROR: BFloat16 is not supported on MPS

before running any Cellpose functions.

[yunchiuuu]

I was running into the same issues (BFloat and numpy version conflicts), turned out my conda was installed as xs64 instead of arm64. I think xs64 pulls the old torch 2.2.2 which still has the BFloat16/MPS bug with cpsam. Check your conda platform first:

conda info | grep platform
# should show osx-arm64

I end up removing the whole anaconda3 and installed the arm64 version of miniforge:

curl -L https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-MacOSX-arm64.sh -o miniforge.sh
bash miniforge.sh

Then:

conda create -n cellpose python=3.11 
conda activate cellpose
pip install cellpose[gui]

End up having numpy=2.4.6 and torch=2.12.0. I've completed a few trains on MPS with no issues now.

[paularandamar]

Thanks for the suggestion. My machine is actually an Intel iMac (x86_64) with an AMD Radeon Pro 580X GPU, not an Apple Silicon Mac. Running uname -m returns x86_64, so osx-64 seems to be the correct platform in my case. From what I understand, the issue may be related to MPS support on Intel Macs rather than an arm64 installation problem. Has anyone successfully run Cellpose 4.x with CPSAM on an Intel Mac using MPS?

[psobolewskiPhD]

Metal Performance Shaders (MPS) backend for GPU acceleration isn't supported on intel macs, see: https://developer.apple.com/metal/pytorch/