[BUG] Cellpose CLI uses CPU when --gpu_device mps is specified

[Kartoffelecke]

Before you fill out this form:

Did you review the FAQ?

• Yes, I reviewed the FAQ on the cellpose ReadTheDocs

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

• Yes, I searched for related previous GH issues.

Now fill this form out completely:

Describe the bug

When I run cellpose from the command line using the recommended setting for a macbook using apple silicon I get the following error

python -m cellpose --dir . --use_gpu --gpu_device mps --save_rois --save_png --diameter 230 --verbose
2026-05-17 23:01:40,118 [INFO] WRITING LOG OUTPUT TO /Users/x/.cellpose/run.log
2026-05-17 23:01:40,118 [INFO]
cellpose version: 	4.1.1
platform:       	darwin
python version: 	3.14.4
torch version:  	2.10.0
2026-05-17 23:01:40,132 [INFO] Neither TORCH CUDA nor MPS version not installed/working.
2026-05-17 23:01:40,132 [INFO] >>>> using CPU
2026-05-17 23:01:40,132 [INFO] >>>> using CPU

Instead when I run the same commands ommiting the --gpu_device mps it uses the GPU and the runtime is 1/5 of before

python -m cellpose --dir . --use_gpu --save_rois --save_png --diameter 230 --verbose
2026-05-17 23:00:51,088 [INFO] WRITING LOG OUTPUT TO /Users/x/.cellpose/run.log
2026-05-17 23:00:51,088 [INFO]
cellpose version: 	4.1.1
platform:       	darwin
python version: 	3.14.4
torch version:  	2.10.0
2026-05-17 23:00:51,104 [INFO] ** TORCH MPS version installed and working. **
2026-05-17 23:00:51,104 [INFO] >>>> using GPU (MPS)
2026-05-17 23:00:51,104 [WARNING] the '--diam_mean' flag is deprecated in v4.0.1+ and no longer used
2026-05-17 23:00:51,106 [INFO] >>>> running cellpose on 18 images using all channels
2026-05-17 23:00:52,002 [INFO] >>>> loading model /Users/x/.cellpose/models/cpsam

To Reproduce

Steps to reproduce the behavior:

1. Install cellpose 4.1.1 using conda
2. Run the above mentioned cellpose CLI commands

Run log

Run logs are included above

[quantixed]

Just to add that I see the same here: --gpu_device mps uses CPU whereas omitting the flag is necessary to use MPS backend. I was messing around with pytorch installation trying to fix the Neither TORCH CUDA nor MPS version not installed/working. error before reading this issue.