Error: _pickle.UnpicklingError: invalid load key, '<'.

[VoHoangAnh]

Thank you for your great work! I set up the environment as requirements.txt, however, I met this error when the model file "data/cvthead.pt" was loaded in inference.py

magic_number = pickle_module.load(f, **pickle_load_args) _pickle.UnpicklingError: invalid load key, '<'

I am very thankful if you can suggest the solution to fix it.

[VoHoangAnh]

I solved it Thank you.

[LeeTeng2001]

hey, what's the solution here???

[VoHoangAnh]

I saved the model file in another PyTorch version.

[cvipym]

Hi, could you please provide more detailed steps on how to save the model file in a different PyTorch version? I'm having trouble understanding the process and would appreciate your guidance.

[VoHoangAnh]

In my case, I used torch 1.11.0 instead of torch 2.0, as illustrated in requirement.txt. you can save the model file in torch 1.11.0 and then use torch 2.0 to load it. I don't know the reason but it worked

[cvipym]

Thank you, I appreciate your response. But, I don't understand saving the model using torch. Aren't files downloaded using bash scripts?

[YacratesWyh]

Google Drive using bash to download would cause viruse scan instead of direct download it, it's an update version. You have to download it manually. If you want a quick check, just check your model, whether they are very tiny like 3KB, then the .sh is failed like me. @HowieMa Hope you fix this

[YacratesWyh]

https://github.com/tensorflow/datasets/issues/3935 To solve this, add like this in bash

[mengsiwei]

Hi, could you please provide more detailed steps on how to save the model file in a different PyTorch version? I'm having trouble understanding the process and would appreciate your guidance.

Hello! Have you solved this problem now? I also encountered this problem.