研究人员开发了MEMAUDIT，一个用于评估LLM代理长期记忆写入能力的创新评估协议。该协议将记忆写入性能与检索和推理分开，从而在预算限制下更精确地分析代理如何将过去的交互压缩成持久记忆。同时，提出了一种名为DARTH-PUM的新型混合处理-使用-记忆架构，它集成了模拟和数字处理-使用-记忆技术，以实现内存阵列内的通用计算。该架构在包括大型语言模型、AES加密和卷积神经网络在内的应用中显示出显著的加速。 AI
影响 新的评估协议和硬件架构可能加速开发更强大、更高效的LLM代理。
排序理由 该集群包含多篇详细介绍LLM新评估协议和硬件架构的研究论文。
AI 生成摘要 · Google Gemini · 来自 4 个来源。 我们如何撰写摘要 →
arXiv:2605.02199v1 Announce Type: new Abstract: Long-term LLM agents must compress streams of past interactions into persistent memory before future queries are known. Existing evaluations usually measure final question-answering accuracy, which entangles memory writing with retr…
arXiv:2602.16075v2 Announce Type: replace-cross Abstract: Analog processing-using-memory (PUM; a.k.a. in-memory computing) makes use of electrical interactions inside memory arrays to perform bulk matrix-vector multiplication (MVM) operations. However, many popular matrix-based k…
arXiv:2604.01707v2 Announce Type: replace Abstract: Memory emerges as the core module in the large language model (LLM)-based agents for long-horizon complex tasks (e.g., multi-turn dialogue, game playing, scientific discovery), where memory can enable knowledge accumulation, ite…
<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@dhodrajsdr192/twenty-repos-one-self-updating-architectural-memory-68678e1af5c9?source=rss------claude-5"><img src="https://cdn-images-1.medium.com/max/1774/1*gd83sQ4lKCdvo6undYO8Yg.png" width=…