About finetune

[morestart]

i use this code load my model

    def configure_model(self) -> None:
        self.model = ModelT(
                img_size=self.cfg.model.img_size,
                patch_size=self.cfg.model.patch_size,
                in_chans=self.cfg.model.in_chans,
                out_chans=self.cfg.model.out_chans,
                embed_dim=self.cfg.model.embed_dim,
                num_groups=self.cfg.model.num_groups,
                num_heads=self.cfg.model.num_heads,
                window_size=self.cfg.model.window_size,
            )
        if self.cfg.trainer.checkpoint_path is not None:
              checkpoint = torch.load(self.cfg.trainer.checkpoint_path, weights_only=False)
              logger.info("Loading pre-trained checkpoint from: %s" % self.cfg.trainer.checkpoint_path)
              state_dict = checkpoint["state_dict"]
              msg = self.model.load_state_dict(state_dict)
              logger.info(msg)

Why does loading the model parameters this way result in a RuntimeError: Error(s) in loading state_dict?
My pretrained model was trained using FSDP.

[morestart]

anyone knows why？