if retrieved_image is None or not isinstance(retrieved_image, torch.Tensor): retrieved_image, mask = load_placeholder_image(self.placeholder_image_path) # Ensure mask has correct format if mask is None: mask
if not add_dummy_prefix: text = "<n>" + text tokens = self._get_text_tokenizer().tokenize(text) return tokens if add_dummy_prefix else tokens[2:] def __getitem__(self, x: Union[int, str]): if isinstance(x, int): if x < self.num_image_tokens: return "<image_{...