How to modify the configuration with cpu?

[rangilyu]

How to modify the configuration with cpu?

[zacario-li]

@lanteng77 You mean train or test?
If test, modify the line from 'cuda:0' to 'cpu'
predictor = Predictor(cfg, args.model, logger, device='cuda:0')

[lanteng77]

I don't have a GPU, so training needs to be modified.

[RangiLyu]

I don't have a GPU, so training needs to be modified.
Sorry, nanodet current not support cpu training. You can try to change all .to('cuda') to .to('cpu') in this project. But I can't make sure that it will work. Just have a try.

[lanteng77]

thk, this is very useful for me. I will try.

[zhongguogu]

(nanodet) gu@local:/home/marchinelearning/nanodet$ python demo/demo.py image --config config/nanodet-m.yml   --model model/nanodet_m.pth  --path input/1.jpg 
model size is  1.0x
init weights...
=> loading pretrained model https://download.pytorch.org/models/shufflenetv2_x1-5666bf0f80.pth
Finish initialize Lite GFL Head.
[root][11-25 22:09:37]INFO:Press "Esc", "q" or "Q" to exit.
Traceback (most recent call last):
  File "demo/demo.py", line 107, in <module>
    main()
  File "demo/demo.py", line 90, in main
    meta, res = predictor.inference(image_name)
  File "demo/demo.py", line 53, in inference
    results = self.model.inference(meta)
  File "/home/marchinelearning/nanodet/nanodet/model/arch/one_stage.py", line 34, in inference
    torch.cuda.synchronize()
  File "/home/gu/anaconda3/envs/nanodet/lib/python3.8/site-packages/torch/cuda/__init__.py", line 378, in synchronize
    _lazy_init()
  File "/home/gu/anaconda3/envs/nanodet/lib/python3.8/site-packages/torch/cuda/__init__.py", line 166, in _lazy_init
    raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

i have modified the line from 'cuda:0' to 'cpu'
predictor = Predictor(cfg, args.model, logger, device='cuda:0') => predictor = Predictor(cfg, args.model, logger, device='cpu'),
but still throw an exception.

[RangiLyu]

(nanodet) gu@local:/home/marchinelearning/nanodet$ python demo/demo.py image --config config/nanodet-m.yml   --model model/nanodet_m.pth  --path input/1.jpg 
model size is  1.0x
init weights...
=> loading pretrained model https://download.pytorch.org/models/shufflenetv2_x1-5666bf0f80.pth
Finish initialize Lite GFL Head.
[root][11-25 22:09:37]INFO:Press "Esc", "q" or "Q" to exit.
Traceback (most recent call last):
  File "demo/demo.py", line 107, in <module>
    main()
  File "demo/demo.py", line 90, in main
    meta, res = predictor.inference(image_name)
  File "demo/demo.py", line 53, in inference
    results = self.model.inference(meta)
  File "/home/marchinelearning/nanodet/nanodet/model/arch/one_stage.py", line 34, in inference
    torch.cuda.synchronize()
  File "/home/gu/anaconda3/envs/nanodet/lib/python3.8/site-packages/torch/cuda/__init__.py", line 378, in synchronize
    _lazy_init()
  File "/home/gu/anaconda3/envs/nanodet/lib/python3.8/site-packages/torch/cuda/__init__.py", line 166, in _lazy_init
    raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

i have modified the line from 'cuda:0' to 'cpu'
predictor = Predictor(cfg, args.model, logger, device='cuda:0') => predictor = Predictor(cfg, args.model, logger, device='cpu'),
but still throw an exception.

Try to comment out all torch.cuda.synchronize().

[siva-wellnesys]

still the issue is same @RangiLyu
i tried these things
1)Try to comment out all torch.cuda.synchronize()
2) have modified the line from 'cuda:0' to 'cpu'
3) changed all .to('cuda') to .to('cpu')

still facing this same issue

File "/home/marchinelearning/nanodet/nanodet/model/arch/one_stage.py", line 34, in inference
torch.cuda.synchronize()
File "/home/gu/anaconda3/envs/nanodet/lib/python3.8/site-packages/torch/cuda/init.py", line 378, in synchronize
_lazy_init()
File "/home/gu/anaconda3/envs/nanodet/lib/python3.8/site-packages/torch/cuda/init.py", line 166, in _lazy_init
raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

is there any alternative for this code to run on CPU?

[wwdok]

@siva-wellnesys There is still a torch.cuda.synchronize() you didn't comment out, this issue should be solved by above discussion.