Efficient-AI-Backbones icon indicating copy to clipboard operation
Efficient-AI-Backbones copied to clipboard

torch.hub ghostnet example is broken due to new cupy dependency

Open NicolasHug opened this issue 3 years ago • 3 comments

Hi all,

We're the maintainers of torch.hub and it looks like executing the code in the ghostnet example fails with the following error:

E             File "python_code/pytorch_vision_ghostnet.py", line 2, in <module>
E               model = torch.hub.load('huawei-noah/ghostnet', 'ghostnet_1x', pretrained=True)
E             File "/home/circleci/miniconda3/lib/python3.8/site-packages/torch/hub.py", line 543, in load
E               model = _load_local(repo_or_dir, model, *args, **kwargs)
E             File "/home/circleci/miniconda3/lib/python3.8/site-packages/torch/hub.py", line 570, in _load_local
E               hub_module = _import_module(MODULE_HUBCONF, hubconf_path)
E             File "/home/circleci/miniconda3/lib/python3.8/site-packages/torch/hub.py", line 88, in _import_module
E               spec.loader.exec_module(module)
E             File "<frozen importlib._bootstrap_external>", line 843, in exec_module
E             File "<frozen importlib._bootstrap>", line 219, in _call_with_frames_removed
E             File "/home/circleci/.cache/torch/hub/huawei-noah_ghostnet_master/hubconf.py", line 6, in <module>
E               from snnmlp_pytorch.models.snn_mlp import SNNMLP
E             File "/home/circleci/.cache/torch/hub/huawei-noah_ghostnet_master/snnmlp_pytorch/models/__init__.py", line 3, in <module>
E               from .snn_mlp import SNNMLP
E             File "/home/circleci/.cache/torch/hub/huawei-noah_ghostnet_master/snnmlp_pytorch/models/snn_mlp.py", line 8, in <module>
E               from .snn_cuda import LIFSpike
E             File "/home/circleci/.cache/torch/hub/huawei-noah_ghostnet_master/snnmlp_pytorch/models/snn_cuda.py", line 10, in <module>
E               import cupy
E           ModuleNotFoundError: No module named 'cupy'

You can find more details about the execution here.

Is cupy now a required dependency to run the ghostnet model? It seems that it wasn't before. I'd just like to make sure whether it's absolutely needed or not, as otherwise we'd need to include it in our CI as well (which we'd preferably not do).

Thanks!

NicolasHug avatar Jan 07 '23 15:01 NicolasHug

'cupy' is a required dependency to run SNN-MLP model in huawei-noah/Efficient-AI-Backbones

iamhankai avatar Jan 11 '23 03:01 iamhankai

Thanks for your reply @iamhankai Could you clarify whether this cupy dependency was added only recently, or it's always been a requirement?

NicolasHug avatar Jan 13 '23 11:01 NicolasHug

@Gaffey

iamhankai avatar Jan 14 '23 11:01 iamhankai