[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-remory-agent-memory":3,"topics-all":38,"news-related-a820bb5a-ac48-4aab-8d8b-34da340900c1":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"a820bb5a-ac48-4aab-8d8b-34da340900c1","REMORY 让 LLM 用 5% 上下文继续跑 Agent","REMORY 用外挂小网络生成「软记忆令牌」补在被压缩摘要后,只占 5.2% 输入位就逼近全上下文得分。Qwen3.8-27B、GLM-5.3-Flash 在 SummHay\u002FBrowseComp\u002FTerminal-Bench 2.1 一致提升,工具调用错误率下降。论文与代码 10 月 8 日同天放出。","长程 Agent 把对话历史压缩进摘要这件事，几家厂商和实验室最近半年已经磨出了好几套方案，包括 AutoCompact、CliffCompaction、CompactionRL 等。但它们都还卡在一道坎上：纯文本摘要保不住每一个后续决策所需的信息，丢掉的东西容易让 Agent 在工具调用、网页检索这类长链路任务里失准。10 月 8 日放出的 arXiv 2610.11287 换了一个思路，引入一个独立的小网络为压缩后的摘要「补」上一段软记忆，把上下文开销砍到原来的 5.2%，却在 SummHay 上逼近全上下文得分。\n\n## 它到底做了什么\n\nREMORY 的核心动作分成两步。第一步，让一个外挂的 memory network 读取完整历史和当前摘要，输出一个固定长度、连续向量化的「软记忆令牌」(soft memory tokens),不是可见文字，而是直接把 LLM 输入嵌入层之外的连续向量接到摘要之后。第二步，主 LLM 冻结不变，只接收「摘要 + 软令牌」即可继续推理，作者把这一段比作「序列维度的残差连接」—— 摘要负责可读叙事的压缩，软记忆负责保留那些不能用文字总结的细节。\n\n和把它当成额外的 KV cache 插入不一样的是「学习」：memory network 的参数是训出来的，输入端永远是冻结的 LLM。研究团队同时为 Qwen3.8-27B 与 GLM-5.3-Flash 各训了一套外挂记忆，分别是 1.9B 与 1.24B 的小模型，权重分别以 `Qwen3.8-27B-REMORY-1.9B`、`GLM-5.3-Flash-REMORY-1.24B` 在 Hugging Face 上公开。\n\n## 测评覆盖哪些场景\n\n论文围绕长程 Agent 跑了四类基准。SummHay 是「带来源归属的高质量摘要」测试，REMORY 在保留 insight coverage 基本不变的前提下，把 source attribution 推高，并能在仅用 5.2% 输入位的情况下逼近 full-context 的 joint score。BrowseComp 与 Terminal-Bench 2.1 是行业流行的网页检索与命令行执行任务，Qwen3.8-27B 和 GLM-5.3-Flash 在两类基准上一致提升，重复工具输出和工具错误的次数显著下降。仓库里的 demo 显示，在 Codex 这类代理里直接插记忆网络的 compact hook 即可上线，Qwen3.8-27B 默认 256K 上下文，配合两卡 80G 显存的 SGLang 部署就能跑。\n\n训练的代价也值得一提：memory network 是有监督学习，目标是让「摘要 + 软令牌」条件下冻结 LLM 输出的 logits 接近「看完整历史」时的输出，因此训练阶段不需要重新训练主模型。这对真正落地至关重要—— 大模型重训代价高、风险大，而 1B 量级的小外挂网络对大部分团队都可承受。\n\n## 行业意义\n\n把上下文从「文本摘要 + 软记忆令牌」双轨压缩，比 KV cache 量化更省显存，比纯文本摘要更准，特别适配 Claude、GPT-6、Qwen3.8-MAX 这类为长程 Agent 准备的指令级模型。REMORY 有点像给 LLM 加了一片「可学习的长期记忆外挂」，和人类工作记忆「读 + 隐式索引」分开的认知结构更接近。一个直接的下一步是把这套思路接到 RAG、Tool-Use 训练管线里，让小模型自己学会压缩、回放工作记忆，而不是每次都让主模型在长上下文里硬扛。","https:\u002F\u002Fgithub.com\u002F1ring2rta\u002FRemory","998df6db-96e6-4b8e-8be1-cfa00a6cd177",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"2d9c2fb0-2be5-4ad1-aedb-e9747addf355","compression",{"id":19,"name":20,"slug":20,"description":14,"color":14},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":22,"name":23,"slug":23,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"7c8f0b26-a86f-4ce8-8e34-b0bbe5e08dc4","en","REMORY: 5.2% residual memory replaces context summaries","REMORY appends learned soft memory tokens after a compacted summary, approaching full-context scores with just 5.2% of input positions on SummHay.","Long-horizon agents have spent the last six months learning to compress their history into summaries. Teams have shipped AutoCompact, CliffCompaction, CompactionRL and a handful of similar systems. They all hit the same wall: a textual summary alone cannot preserve every detail a downstream decision depends on, and the lost information shows up as repeated tool calls, broken web search and broken shell execution. The arXiv paper 2610.11287, posted on October 8, takes a different angle. It trains a small external network to append learned soft memory tokens to the compacted summary, cutting context usage to 5.2% of the original while approaching the full-context score on SummHay.\n\n## What REMORY actually does\n\nThe pipeline has two steps. First, an external memory network reads the full history and the current summary and outputs a short sequence of continuous soft memory tokens. These tokens are not visible text but continuous vectors that the main LLM receives outside its input embedding layer. Second, the main LLM stays frozen and only sees \"summary + soft tokens\"; the authors describe this layer as a residual connection along the sequence dimension, where the summary carries the readable narrative and the soft tokens carry the rest.\n\nThe trick versus pushing more KV cache into the prompt is that the memory network is trained. Its loss targets the logits of the frozen LLM conditioned on \"summary + soft tokens\" matching the logits when the same frozen LLM sees the full history. As a result, the main model never needs to be retrained. The team shipped separate memory checkpoints for Qwen3.8-27B and GLM-5.3-Flash: a 1.9B and a 1.24B parameter network, released on Hugging Face as `Qwen3.8-27B-REMORY-1.9B` and `GLM-5.3-Flash-REMORY-1.24B` respectively.\n\n## What the benchmarks cover\n\nThe paper evaluates four benchmarks aimed at long-horizon agents. On SummHay, a summarization benchmark that scores both source attribution and insight coverage, REMORY pushes source attribution up at near-unchanged insight coverage and approaches the full-context joint score using only 5.2% of the input positions. BrowseComp and Terminal-Bench 2.1, the popular web search and shell execution tasks, both see consistent gains on Qwen3.8-27B and GLM-5.3-Flash, with substantially fewer repeated tool outputs and tool errors. The repository ships Codex-style demos where the memory network's compact hook plugs straight into an existing agent harness, and Qwen3.8-27B runs at its default 256K context on two 80 GB GPUs through SGLang.\n\nThe training cost is the practical hook. Because the supervision targets the frozen LLM's logits, only the 1B-scale memory network is updated, which fits the budget of most teams that cannot afford to retrain a 27B or larger model.\n\n## What this means for the field\n\nCombining textual summarization with learned soft memory uses less memory than KV cache quantization and beats pure text summarization on long-horizon tasks. It is a clean fit for frontier instruction-tuned models built for agents, including Claude, GPT-6 and Qwen3.8-MAX. Structurally REMORY behaves like an external learned long-term memory module, and the parallel to human working memory, where readable recall and implicit indexing are separated, is worth keeping in mind. The obvious next step is to wire this idea into RAG and tool-use training pipelines so small models learn to compress and replay their own working memory instead of forcing the main model to carry long contexts on every turn.","remory-agent-memory","2026-10-11T09:00:00Z","2026-10-11T05:13:30.665267Z","2026-10-11T05:13:30.665277Z",true,"agent",84,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"dab2e487-cfba-4d75-a93f-eff133476348","剪枝校准对不上解码分布?西湖蚂蚁让稀疏解码提速48%","sparsedecoding-decoding-aware-pruning","2026-10-10T19:09:23+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"feb04a1c-2c75-42d2-9b62-9c4f0585d924","AutoCompact:让编码代理自己学会何时压缩上下文","autocompact-learned-context-compaction","2026-10-04T17:10:39+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"88d15f9c-5548-4dae-bde5-75e79c3d0a28","8K 上下文顶 1M:把长文档问题训练成 agent 接力,而不是堆窗口","trained-agentic-context-management","2026-10-01T11:32:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"5fa17d26-6090-4805-893a-aa880cf33369","让 Agent 学会遗忘:删掉旧推理,分数反而涨了","agents-forget-reasoning-iclr-compression","2026-09-28T19:30:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"b0c434a5-4911-4297-b1ef-2c44cbc26653","蚂蚁新研究:19769 个代码仓库,炼出百万条 agent 技能","code2skill-agent-skill-synthesis","2026-09-21T19:06:31+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"b0c4e8d2-5662-4e3e-b489-6202eabbe97b","Dream-RSI 把历史当模拟器:162 倍杠杆重写 RSI 算力账本","dream-rsi-replay-simulator-162x","2026-09-16T06:00:00+00:00"]