As large language models (LLMs) advance toward increasingly long inputs, soft context compression offers a promising approach to efficient inference by conve…
机构:中科院
来源:arXiv 2611.16384 | AI4Papers 论文推荐平台