Soft context compression condenses a context into a few memory tokens that a frozen LLM consumes in place of the raw text, but existing compressors fix the c…
机构:东京大学
来源:arXiv 2609.11192 | AI4Papers 论文推荐平台