11-20 更新 实时
首页/ 公司动向/ 正文

Meta FAIR 自改进记忆持续学习得分 56.0

公司动向 来源:MarkTechPost 2026-11-20 05:46 1 阅读 机器翻译

Meta FAIR 推出了 Self-Improving Memory,这是一种让 AI 智能体自行管理记忆的方法。模型首先自行发现自己的记忆指令,随后将这些技能训练进自身权重中。共同作者包括来自华盛顿大学和约翰斯·霍普金斯大学的研究人员。

Self-Improving Memory 是 Meta FAIR 提出的一种两阶段方法,用于教会智能体模型管理基于文件的记忆。智能体连续处理一系列子任务,每个子任务都从空上下文开始,只有通过文件工具编辑的持久化记忆目录会被保留下来。

两个提示词控制着这一记忆:OFFLOAD 在子任务结束后更新记忆,LOAD 则在下一个子任务开始前检索笔记。通用版本把策略留给模型自行决定。研究博文中一项先导研究发现,没有任何一种人工编写的策略能在所有任务上胜出。

在第一阶段,一个元智能体重写 LOAD 和 OFFLOAD 提示词。搜索从 4 个人工编写的种子提示词出发:Summary、Retrieval、Skill 和 Hypothesis。每个候选提示词都会完整跑 5 遍任务流,按平均得分排名前 4 的成为修订的父代。

对于每个父代,元智能体会阅读轨迹、得分和最终的记忆目录,然后提出修订后的提示词。筛选使用一个元智能体无法看到的留出集。

发现的提示词非常具体。其中一个 Database 提示词要求在回答问题前先读取 GUARDRAILS.md,再读取 SCHEMA.md。最佳的 Codebase 设计使用一个快速入门文件、一个带索引的修复注册表以及独立的修复文件。

第二阶段遵循“发现—演示—蒸馏”流程。任务智能体使用优化后的提示词运行训练子任务,每一次模型调用都会被记录。随后,研究团队将这段历史中的优化提示词换回通用提示词。

这种替换之所以有效,是因为该行为在通用指令下依然成立。示例包括保留 QUICKSTART.md、REGISTRY.md 和 SCHEMA.md 文件,以及在聚合前校验连接键。模型使用 LoRA、秩为 64,在注意力投影层上进行微调。损失只覆盖助手 Token,包括推理过程和工具调用。每个任务权重相同。

元智能体和任务智能体是同一个 Kimi-K2.6 模型。因此,模型是在它自己发现的记忆技能上训练的。

测试使用 3 个 Continual Learning Bench 任务:Database(SQL)、Codebase(代码库修复)和 Spectrum(无线电扫描)。测试子任务遵循设计好的相位变化。因此,结果衡量的是对同一任务新实例的泛化能力。

来源:Self-Improving Memory 博客的图 4 和图 5。所有行均为 Kimi-K2.6。

优化后的 harness 在全部 3 个任务上都击败了最佳手写种子。Codebase 提升最大,达 20.6 分。搜索过程噪声较大:44 个生成的修订版本中有 37 个未能超越此前的最佳成绩。Database 和 Spectrum 在第一代就达到峰值。Codebase 则一直提升到第 4 代。

蒸馏之后,每个模型都运行通用 harness。蒸馏后的模型在 10 个匹配种子上的平均分为 43.1。在通用轨迹上做 SFT 达到 37.6。这相当于 5.5 分的提升,即 14.6%。

近期有若干方法在智能体记忆或 harness 上进行搜索。Self-Improving Memory 搜索记忆提示词,再将其蒸馏进权重。由于基准测试不同,标称数字无法直接比较。

Harness-Zero 最为接近,但它使用一个独立的智能体来纠正学生模型。Self-Improving Memory 跳过了这一步。

欢迎查看这篇博客文章、X 上的推文串以及 Continual Learning Bench 论文。你可以通过我们的 GPULeaderboard.com Chrome 扩展程序在此查看部署方案和规格。

需要与我们合作来推广你的 GitHub 仓库、Hugging Face 页面、产品发布或网络研讨会等?请联系我们。

文章《Meta FAIR Self-Improving Memory 在 Continual Learning Bench 上达到 56.0》最早发布于 MarkTechPost。

译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(Meta FAIR Self-Improving Memory Hits 56.0 on Continual Learning Bench)

Meta FAIR has introduced Self-Improving Memory, a method for AI agents that manage their own memory. The model first discovers its own memory instructions. It then trains those skills into its weights. Co-authors include researchers from the University of Washington and Johns Hopkins University.

TL;DR

  • Method: The base is Kimi-K2.6, a 1T-parameter MoE with 32B active and 256K context. Distillation adds a LoRA adapter (rank 64) on attention projections.
  • Runs on: Training hardware is not disclosed. Kimi-K2.6 serving needs a multi-GPU node; Vast.ai recommends 8x H200. Native INT4 weights exist.
  • Performance: Self-found memory prompts average 56.0 on 3 tasks, versus 44.6 for the best human-written ones.
  • Best: Codebase, 64.3 with optimized prompts, +20.6 over the best hand-written harness.
  • Worst: Spectrum distillation adds only 2.9 points (46.5 vs 43.6 for SFT on generic runs).

What is Self-Improving Memory?

Self-Improving Memory is a 2-stage method from Meta FAIR that teaches an agent model to manage file-based memory. The agent works through a stream of subtasks. Each subtask starts with an empty context. Only a persistent memory directory, edited with file tools, carries over.

Two prompts control that memory. OFFLOAD updates memory after a subtask. LOAD retrieves notes before the next one. Generic versions leave the strategy to the model. A pilot study in the research post found that no single hand-written strategy wins every task.

How does harness optimization work?

In stage 1, a meta agent rewrites the LOAD and OFFLOAD prompts. The search starts from 4 human-written seeds: Summary, Retrieval, Skill and Hypothesis. Each candidate runs the full task stream 5 times. The top 4 by mean score become parents for revision.

For each parent, the meta agent reads trajectories, scores and final memory directories. It then proposes revised prompts. Selection uses a holdout set the meta agent cannot see.

The discovered prompts are concrete. One Database prompt says to read GUARDRAILS.md, then SCHEMA.md, before answering. The best Codebase design uses a quickstart file, an indexed fix registry and separate fix files.

How does harness distillation move the gains into weights?

Stage 2 follows a Discover-Demonstrate-Distill pipeline. The task agent runs train subtasks with the optimized prompts. Every model call is logged. The research team then swaps the optimized prompts back to the generic ones in that history.

The swap works because the behavior stays valid under generic instructions. Examples include keeping QUICKSTART.md, REGISTRY.md and SCHEMA.md files, and verifying join keys before aggregation. The model is fine-tuned with LoRA, rank 64, on attention projection layers. The loss covers only assistant tokens, including reasoning and tool calls. Each task gets equal weight.

The meta agent and task agent are the same Kimi-K2.6 model. So the model is trained on memory skills it discovered itself.

What do the benchmarks show?

The tests use 3 Continual Learning Bench tasks: Database (SQL), Codebase (repo fixes) and Spectrum (radio scans). Test subtasks follow a designed phase shift. So results measure generalization to new instances of the same tasks.

Test scoreDatabaseCodebaseSpectrumAverage
No memory7.531.821.920.4
Compaction39.836.934.637.1
Generic harness32.333.542.536.1
Best human-written47.943.742.144.6
Optimized harness53.864.349.956.0
SFT on generic trajectories31.637.543.637.6
SFT on optimized trajectories (generic prompts)37.745.146.543.1

Source: Figures 4 and 5, Self-Improving Memory blog. Kimi-K2.6 for all rows.

Optimized harnesses beat the best hand-written seed on all 3 tasks. Codebase gained the most, at 20.6 points. Search was noisy: 37 of 44 generated revisions failed to beat the previous best. Database and Spectrum peaked in the first generation. Codebase kept improving through generation 4.

After distillation, every model runs the generic harness. The distilled model averages 43.1 across 10 matched seeds. SFT on generic trajectories reaches 37.6. That is a 5.5-point gain, or 14.6%.

How does it compare with ALMA, MemEvolve and Harness-Zero?

Several recent methods search over agent memory or harnesses. Self-Improving Memory searches memory prompts, then distills them into weights. Benchmarks differ, so headline numbers are not directly comparable.

FeatureSelf-Improving MemoryALMAMemEvolveHarness-Zero
OrgFAIR at Meta, UW, JHUXiong, Hu, CluneOPPO AI Agent Team, LV-NUSPeking University, Google, HKUST
What is searchedLOAD/OFFLOAD memory promptsMemory designs as Python codeMemory architecture and contentsFull agent harness
Trains weightsYes, LoRA rank 64NoNoYes, LoRA SFT
ModelsKimi-K2.6 (agent and meta agent)GPT-5 meta agent, GPT-5-nano/mini task agentGPT-5-mini (plus DeepSeek V3.2, Kimi K2)Qwen3.5-9B student
BenchmarksContinual Learning Bench (3 tasks)ALFWorld, TextWorld, Baba Is AI, MiniHackGAIA, WebWalkerQA, xBench-DeepSearch, TaskCraftSpreadsheetBench Verified, AppWorld, USPTO Retrosynthesis
Headline result56.0 vs 44.6 best human harness; 43.1 distilled53.9% vs 41.1% no memory (GPT-5-mini)Up to 17.06% gains23.3% to 44.3% with harness removed
CodeNot releasedGitHubGitHubGitHub

Harness-Zero is the closest match, but it uses a separate agent to correct the student. Self-Improving Memory skips that step.

Key Takeaways

  • Kimi-K2.6 writes memory prompts that beat human ones: 56.0 vs 44.6.
  • Codebase sees the biggest jump: 64.3, up 20.6 points.
  • Distillation keeps part of the gain: 43.1 vs 37.6 with generic prompts.
  • 37 of 44 meta-agent revisions failed, so evaluation and selection matter.


Check out the blog post, the X thread and the Continual Learning Bench paper. You can check the deployment options and specs here through our GPULeaderboard.com chrome extension.

Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.

Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us

The post Meta FAIR Self-Improving Memory Hits 56.0 on Continual Learning Bench appeared first on MarkTechPost.

正文保留英文原文,标题与摘要由机器翻译。

本文由采集器自动同步

原文发布于 MarkTechPost(Meta FAIR Self-Improving Memory Hits 56.0 on Continual Learning Bench),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗