読書リサーチレーダー投資フレームワーク
ログイン / 新規登録
← 読解に戻る
ニューストピック · CoEM延迟压缩记忆クイックテイク1 分で読めるDisconfirmAI 編集部

Delayed-compression memory CoEM lifts long-context reasoning F1 by over 10 points in self-tests

CoEM postpones compression until later context confirms an excerpt's relevance; authors' own tests show 10.4-11.4 F1 gains over the strongest baseline on Qwen3.5-9B, pending independent replication.

2026-09-30 12:00 GMT+8

Readers can now learn of CoEM, a memory-management method for long-context reasoning: potentially useful source excerpts are kept verbatim in a pending set, and only after later context clarifies their relevance does the method commit them to compressed memory or discard them, avoiding the loss of key details from premature compression; the authors' own tests show F1 gains of more than 10 points.

The prior practice was to compress memory early, which could discard key details before an excerpt's relevance became clear, causing information loss in long-context reasoning.

The decision policy is trained with reinforcement learning, and a frozen verifier accepts memory facts only when supported by retained excerpts. On 6,400-document inputs, the authors' self-tests show CoEM beats the strongest memory baseline by 10.4-11.4 F1 points on Qwen3.5-9B.

The experiments cover a single model, Qwen3.5-9B, and a single input length of 6,400 documents; performance on other models and shorter or longer contexts is unknown, and no independent replication exists yet. The code is open on GitHub for verification.

ソース

  • arxiv.org↗
トピック

CoEM延迟压缩记忆

ログイン / 新規登録
読書リサーチレーダー投資フレームワーク

CoEM(Commit-on-Evidence Memory)是 KAIST、澳门大学、MIT 等机构作者提出的一种长文本推理记忆框架,核心是把"压缩"这个不可逆决定推迟:模型分块读长文档时,暂不确定是否有用的原文片段先逐字存入待定区,等后…

4 ステップ1 件更新日 2026-10-03 20:48

経緯

  1. 2026-09-29论文 v1 提交 arXiv(cs.AI),提出 CoEM:在固定 1,024 token 记忆预算(256 待定 + 768 已提交)下延迟压缩证据,用强化学习结合最终答案奖励与步级证据奖励训练记忆策略。arxiv.org ↗
  2. 2026-09-30论文更新至 v2。arxiv.org ↗
  3. 2026-09-30GitHub 仓库上线项目概览与文档,宣布完整源码将在论文接收后发布于 GitHub、模型权重发布于 Hugging Face;仓库当前仅含文档与图表。github.com ↗

未解決の問題

  • 10.4–11.4 F1 的提升能否被独立团队复现,尤其是在其他骨干模型上?
  • 论文能否通过同行评审并被会议接收,从而触发完整源码与权重的发布?
  • 延迟保留待定证据在真实生产场景(而非三个多跳 QA 基准)中的计算开销与收益如何?
トピック全文を読む →
リサーチを続ける →