torch.load raise TypeError: 'strict' is an invalid keyword argument for Unpickler()

[wwq66]

raise TypeError: 'strict' is an invalid keyword argument for Unpickler() when running the code below. It seems torch.load does not have the keyword 'strict'.

def _load_checkpoint(self, url_or_filename):
        """
        Resume from a checkpoint.
        """
        if is_url(url_or_filename):
            cached_file = download_cached_file(url_or_filename, check_hash=False, progress=True)
            checkpoint = torch.load(cached_file, map_location=self.device, strict=False)
        elif os.path.isfile(url_or_filename):
            print("url_or_filename: ", url_or_filename)
            checkpoint = torch.load(url_or_filename, map_location=self.device, strict=False)
        else:
            raise RuntimeError("checkpoint url or path is invalid")

[RenShuhuai-Andy]

Hi, what's your torch version? You can try pip install torch==2.1.1 torchvision==0.16.1 torchaudio==2.1.1 --index-url https://download.pytorch.org/whl/cu118

[wwq66]

I've tried torch version of 1.12.1, 1.13.1, 2.1.0, but none of these version works. And the pytorch docs shows that torch.load do not have the keyword 'strict' as below.

Only the load_state_dict have the 'strict' keyword

[RenShuhuai-Andy]

I see. I think you can directly remove the argument of strict=False. I will fix it later.

[wwq66]

Removing strict can actually fix the error. But seems to raise another error about missing keys while loading state_dict. I'll close this issue temporarily and try to fix the missing keys error.