Fix on base_model.py: load_checkpoint_from_config(), fix the bug of loading pretrained model

[neumyor]

I was loading pretrained model and made a mistake in my config, then I found this funny bug in models/base_model.py.

fix it with a simple assert pretrain_path is not None, "Found load_finetuned is False, but pretrain_path is None."

def load_checkpoint_from_config(self, cfg, **kwargs):
        """
        Load checkpoint as specified in the config file.

        If load_finetuned is True, load the finetuned model; otherwise, load the pretrained model.
        When loading the pretrained model, each task-specific architecture may define their
        own load_from_pretrained() method.
        """
        load_finetuned = cfg.get("load_finetuned", True)
        if load_finetuned:
            finetune_path = cfg.get("finetuned", None)
            assert (
                finetune_path is not None
            ), "Found load_finetuned is True, but finetune_path is None."
            self.load_checkpoint(url_or_filename=finetune_path)
        else:
            load_pretrained = cfg.get("load_pretrained", True)
            if load_pretrained:
                # load pre-trained weights
                pretrain_path = cfg.get("pretrained", None)
                assert "Found load_finetuned is False, but pretrain_path is None."  # this is the bug, lol
                self.load_from_pretrained(url_or_filename=pretrain_path, **kwargs)

[salesforce-cla]

Thanks for the contribution! Before we can merge this, we need @neumyor to sign the Salesforce Inc. Contributor License Agreement.

[neumyor]

Already signed the cla, pls have a review @dxli94