Through pre-training on extensive text and image datasets, current multi-modal large language models (MLLMs) achieve strong performance on general tasks. How…
机构:浙大
来源:arXiv 2609.15668 | AI4Papers 论文推荐平台