Illegal instruction

[zoldaten]

hi!
i try to inference on raspberry pi:
python3 predict_text_prompt.py --source ultralytics/assets/bus.jpg --checkpoint pretrain/yoloe-v8s-seg.pt --names person
but got error.
the problem with this string in predict_text_prompt.py:
model.set_classes(args.names, model.get_text_pe(args.names))
in cmd it returns:

Traceback (most recent call last):
  File "/home/pi/yoloe/predict_text_prompt.py", line 89, in <module>
    main()
  File "/home/pi/yoloe/predict_text_prompt.py", line 56, in main
    model.set_classes(args.names, model.get_text_pe(args.names))
  File "/home/pi/yoloe/ultralytics/models/yolo/model.py", line 99, in get_text_pe
    return self.model.get_text_pe(texts)
  File "/home/pi/yoloe/venv/lib/python3.9/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
  File "/home/pi/yoloe/ultralytics/nn/tasks.py", line 626, in get_text_pe
    assert(not self.training)
AssertionError

how to fix ?

[jameslahm]

Thanks for your interest! We cannot reproduce the issue in our environment. Could you please try add model.eval() after this line? Thanks!

yoloe/predict_text_prompt.py

Line 52 in a803c8b

model.to(args.device)

[zoldaten]

Thanks for your interest! We cannot reproduce the issue in our environment. Could you please try add model.eval() after this line? Thanks!

yoloe/predict_text_prompt.py

Line 52 in a803c8b

model.to(args.device)

not working.
there is something with Build text model mobileclip:blt
it works on Windows. so that i just pickle dump pickle.dump(model.get_text_pe(args.names),f) on windows, copy-paste
and pickle load on raspberry and it starts working. but it is 'handmade fix'.

[zoldaten]

ok, i fix it. that was about mobileclip
steps as follows:

1. make venv only for mobileclip (do not install yoloe before)

git clone https://github.com/apple/ml-mobileclip.git
cd ml-mobileclip
python3 -m venv venv
source venv/bin/activate
pip install -U pip
pip install -e .

2. install yoloe

cd ~
git clone https://github.com/THU-MIG/yoloe.git

nano requerements.txt

*comment there:
#-e third_party/ml-mobileclip

pip install -e .

3.upgrade torch and torchvision if needed:
pip3 install torch==2.6 torchvision==0.21