janus/janusflow/models/processing_vlm.py [407:420]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
            seq_lens.append(len(prepare))

        input_token_max_len = max(seq_lens)
        max_n_images = max(1, max(n_images))

        batched_input_ids = torch.full(
            (batch_size, input_token_max_len), self.pad_id
        ).long()  # FIXME
        batched_attention_mask = torch.zeros((batch_size, input_token_max_len)).long()
        batched_pixel_values = torch.zeros(
            (batch_size, max_n_images, *self.image_processor.default_shape)
        ).float()
        batched_images_seq_mask = torch.zeros((batch_size, input_token_max_len)).bool()
        batched_images_emb_mask = torch.zeros(
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



janus/models/processing_vlm.py [376:389]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
            seq_lens.append(len(prepare))

        input_token_max_len = max(seq_lens)
        max_n_images = max(1, max(n_images))

        batched_input_ids = torch.full(
            (batch_size, input_token_max_len), self.pad_id
        ).long()  # FIXME
        batched_attention_mask = torch.zeros((batch_size, input_token_max_len)).long()
        batched_pixel_values = torch.zeros(
            (batch_size, max_n_images, *self.image_processor.default_shape)
        ).float()
        batched_images_seq_mask = torch.zeros((batch_size, input_token_max_len)).bool()
        batched_images_emb_mask = torch.zeros(
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -