后训练自动化:AIBuildAI 用元搜索在 PostTrainBench 拿第一,46.6 分超过所有 Agent
Autonomous LLM Post-Training: AIBuildAI's Meta Search Tops PostTrainBench at 46.6, Beating Every Agent
📌 开源仓库:https://github.com/aibuildai-inc/aibuildai-llm-posttrain-agent 技术报告:github.com/aibuildai-inc/aibuildai-llm-posttrain-agent/blob/main/docs/aibuildai-llm-post-train-agent.pdf License:Apache 2.0 | Stars:7 | Language:Python | 联系:[email protected]
后训练为什么难?
预训练好的语言模型不能直接用。Post-training(后训练)是让模型能听指令、能推理、会拒绝有害请求的那个阶段,它决定了模型在实际使用中是否有用。
但后训练的流水线设计成本高、门槛高。主要原因是决策之间强耦合:
- 用哪个偏好对齐算法(SFT → DPO / GRPO / PPO),取决于数据混合比
- 数据混合比,取决于基座模型的特点
- 学习率调度,取决于算法选择
- 任何一个错,要从头重跑——H100 上一次完整训练就要花几小时
PostTrainBench 把这个问题形式化成了一个基准测试:给定基座模型 + 目标能力,Agent 必须在十小时预算内自主完成整个后训练流水线,结果由基准自己的隐藏测试集评分。
AIBuildAI 的方案:元搜索(Meta Search)
以往的思路是把 Claude Code 或 Codex 这类通用编程 Agent 对准任务跑。这能出结果,但有一个根本限制:Agent 在早期就提交了一个方案,后续所有决策都沿着这条线走,一旦卡住只会在同一条线上优化,不会放弃再换一个方向。
AIBuildAI 的核心贡献是元搜索——让搜索拓扑本身成为 Agent 的输出。
传统树搜索(每个节点是一次完整训练实验):
- 所有任务走同一种树形结构
- 无法表达「多阶段课程」「基于奖励模型的在线 RL」「多模型投票」等结构
- 数据生成阶段与训练阶段捆在一起,不能复用前缀
元搜索:
- 元 Agent 先读任务、查知识库,然后写一个搜索程序
- 搜索程序由三种原语组成:
Agent(带工具的 LLM 会话)、Program(确定性计算)、Search(组合编排) - 可以生成 DAG、锦标赛、多路融合、迭代循环——不限于树
- 支持「运行时重规划」:一阶段结束后,剩余预算和上一阶段结果交给新的元 Agent,它根据实测结果决定下一阶段的结构
每次运行的拓扑都是任务专属的,而不是固定预设。
知识系统:四个语料库通过 MCP 检索
元 Agent 和实验 Agent 在做设计决策时,都能查询一个远程检索服务(Kb),通过 MCP 接口访问,按语义嵌入检索。
四个子语料库:
| 类别 | 规模 | 内容 |
|---|---|---|
| 工作流(Workflow) | 1 个技能文档,34 张参考卡 | 后训练流程的 13 个顺序研究步骤,每步说明要做的判断、可选方向、不可迁移的情况 |
| 方法论(Methodology) | 104 张方法卡,30 个有代码实现 | 每张卡含原始论文、数学推导、数值样例、各库默认参数、理论与实际成本、训练信号监控 |
| 数据集(Dataset) | 215 张数据集卡,13 个推荐,34 个须隔离 | 每张卡含许可证、列定义、版本锁定加载代码、真实样本行、采纳证据、筛查记录 |
| 框架(Framework) | 108 张库卡,16 个推荐 | 何时选这个库、如何启动训练、监控什么指标、如何保存可被加载的结果 |
每张技能文档同时记录「什么在什么条件下有效」和「什么在什么条件下失败」,Agent 可以从正负两个方向检索。
PostTrainBench 结果
七个任务:AIME 2025(竞赛数学)、GSM8K(数学)、ArenaHard Writing(创意写作)、GPQA Main(问答)、HealthBench(医疗建议)、HumanEval(代码生成)、BFCL(函数调用)
四个基座模型:Qwen3-1.7B-Base、Qwen3-4B-Base、SmolLM3-3B-Base、gemma-3-4b-pt
每次运行:单张 H100 GPU,10 小时上限,全自主,无人工介入,所有 Agent 角色跑在 Claude Opus 5 上
| 方法 | AIME | ArenaHard | BFCL | GPQA | GSM8K | HealthBench | HumanEval | 总分 |
|---|---|---|---|---|---|---|---|---|
| AIBuildAI(本文) | 15.8 | 65.5 | 95.8 | 31.4 | 82.3 | 42.6 | 69.0 | 46.6 |
| Locus (Opus 5) | 9.4 | 66.3 | 94.3 | 33.1 | 82.9 | 44.3 | 66.6 | 45.6 |
| Claude Code (Fable 5) | 13.3 | 61.5 | 73.0 | 28.3 | 83.5 | 40.1 | 58.4 | 41.8 |
| Claude Code (Opus 5) | 9.2 | 53.4 | 1.5 | 31.4 | 80.3 | 38.7 | 59.8 | 35.0 |
| Official instruct(参考基线) | 29.2 | 70.2 | 85.0 | 36.2 | 87.0 | 43.3 | 71.5 | 51.1 |
每格为四个基座模型的平均值;加粗为 Agent 类最高
三个领先任务的原因分析(来自报告):
BFCL(函数调用,95.8):任务核心是让模型输出符合 schema 的函数调用格式,是一个「输出契约」问题——学会了契约就接近满分,学不会就接近零分。元搜索能精准定位并针对这个契约设计流水线,而通用 Agent 容易绕开这一步。
HumanEval(代码生成,69.0):在全部四个基座模型上都领先。代码任务有明确的自动化验收标准(单元测试通过率),奖励信号清晰,适合迭代优化。
AIME 2025(竞赛数学,15.8):数学推理对格式要求严格(整数输出),且 SFT + GRPO 的叠加策略在小模型上效果差异很大。元搜索能根据中间结果调整策略,而固定拓扑在早期做了错误决策后难以纠正。
唯一在前面的是 Official instruct(51.1)——这是各模型官方开发团队发布的指令微调版本,属于「参考基线」而非竞争 Agent,使用的算力和时间预算不受约束。
和竞争方案的核心区别
Locus(外部提交,Intology,也跑在 Opus 5):总分 45.6,在 ArenaHard、GPQA、HealthBench 三个任务上领先。根据报告描述,Locus 是目前外部提交中最强的单一 Agent 方案。AIBuildAI 在这三个任务上与它的差距都在 2 分以内。
Claude Code + Fable 5(41.8):Claude Code 在 BFCL 上出现了 73.0 ± 28.0 的高标准差,说明在函数调用任务上不稳定。这与前文的分析一致——Claude Code 是通用代码 Agent,不针对后训练知识设计。
Claude Code + Opus 5(35.0):BFCL 只有 1.5 分,是明显的格式输出失败——说明同一模型(Opus 5)在不同架构下差距巨大。知识系统 + 元搜索的组合在这里发挥了关键作用。
技术细节:三个用树搜索无法表达的场景
报告明确列举了为什么树搜索不够,这三个场景展示了元搜索的必要性:
1. 数据生成阶段 树节点必须是已训练并评分的 pipeline,但数据语料只有在模型训练完成后才能评分。要比较 k 种数据生成方案,树搜索必须跑 k 次完整训练。元搜索可以把数据生成作为独立阶段,用廉价的代理指标(验证器通过率、去重率)筛选语料,再把好的语料传入训练阶段。
2. 多阶段课程学习 SFT → 偏好对齐 → 拒绝采样微调(Rejection Fine-Tuning)的顺序未知。树搜索每个候选都要从头训练共享前缀。元搜索可以缓存前缀,只搜索后续阶段。
3. 在线强化学习 奖励模型和策略模型需要在一次连续训练中交替更新(策略更新 → 奖励模型重拟合 → 继续训练)。树搜索只能在训练完成后评分,无法在训练中介入。元搜索可以把这表达成一个迭代循环。
产品定位与局限
AIBuildAI 的市场定位是:帮企业把专有数据转化为高性能定制 LLM,降低人力、时间和试错成本。
从技术报告来看,当前系统的实际限制:
- 需要 Linux x86_64 + systemd + cgroup-v2,不能跑在 macOS 上
- 每次运行需要单张 H100(十小时预算),硬件成本不低
- 所有 Agent 运行在 Claude Opus 5 上,API 成本叠加在训练成本上
- 仓库刚开放(2026-09-16),Stars 只有 7,代码成熟度待验证
- PostTrainBench 上的结果是「当前快照」,报告明确说系统仍在开发中
技术路线本身是清晰的:元搜索 + 领域知识库 + 自动化实验 = 比通用 Agent 更高效的后训练探索。这与「人意图 + Agent 执行」的当前 AI 工程方向高度一致。
开源代码与模型仅供学习、勿直接用于工作。
© 2026 Author: Mycelium Protocol. 本文采用 CC BY 4.0 授权——欢迎转载和引用,须注明作者姓名及原文链接,不得去除署名后以原创发布。
📌 Repository: https://github.com/aibuildai-inc/aibuildai-llm-posttrain-agent Technical report: github.com/aibuildai-inc/aibuildai-llm-posttrain-agent/blob/main/docs/aibuildai-llm-post-train-agent.pdf License: Apache 2.0 | Stars: 7 | Contact: [email protected]
Why Is Post-Training Hard?
A pretrained language model isn’t ready to use. Post-training is the stage that makes a model follow instructions, reason about its outputs, and decline harmful requests — it determines whether a model is actually useful in practice.
But designing post-training pipelines is expensive and expertise-intensive. The core reason: decisions are tightly coupled.
- Which preference alignment algorithm to use (SFT → DPO / GRPO / PPO) depends on the data mixture
- The effective data mixture depends on the base model’s characteristics
- The learning rate schedule depends on the algorithm choice
- Any mistake means a full rerun — one complete training run on an H100 takes hours
PostTrainBench formalizes this as a benchmark: given a base model and a target capability, an agent must autonomously complete the entire post-training pipeline within a ten-hour budget, scored by the benchmark’s own held-out evaluation.
AIBuildAI’s Approach: Meta Search
The prior approach was to point a general coding agent (Claude Code, Codex) at the task and let it run. This produces results, but has a fundamental limitation: the agent commits to one approach early and every subsequent decision is conditioned on its existing trajectory — when it stalls, it refines the same line rather than abandoning it for something different.
AIBuildAI’s core contribution is meta search: making the search topology itself the output of an agent.
Traditional tree search (each node is one complete training experiment):
- Every task goes through the same tree structure
- Cannot express multi-stage curricula, online RL against a learned reward model, multi-model voting
- Data generation and training are bundled — no prefix caching
Meta search:
- A meta agent reads the task, consults the knowledge base, then writes a search program
- The program is composed from three primitives:
Agent(LLM session with tools),Program(deterministic computation),Search(composite orchestration) - Can generate DAGs, tournaments, fan-ins, iterative loops — not just trees
- Supports “runtime replanning”: at each stage boundary, the remaining budget and stage results are handed to a new meta agent, which decides the next stage’s structure based on measured outcomes
Each run’s topology is task-specific rather than fixed in advance.
Knowledge System: Four Corpora via MCP
Both the meta agent and experimenter agents query a remote retrieval service (Kb) via MCP, using semantic embedding search when making design decisions.
| Corpus | Scale | Coverage |
|---|---|---|
| Workflow | 1 skill doc, 34 reference cards | 13 ordered research actions per run, each explaining the judgment it settles, options available, and non-transferable cases |
| Methodology | 104 method cards, 30 with code | Defining paper, math, numerical example, per-library defaults, theory/practice cost, signals to watch during training |
| Dataset | 215 cards, 13 recommended, 34 held-out | License, column definitions, version-pinned load line, real sample row, adoption evidence, screening record |
| Framework | 108 library cards, 16 recommended | When to pick, how to start, what to monitor, how to save a loadable result |
Each skill document records both what works under what conditions, and what fails and why — both positive and negative evidence are available.
PostTrainBench Results
Seven tasks: AIME 2025 (competition math), GSM8K (math), ArenaHard Writing (creative writing), GPQA Main (QA), HealthBench (health advice), HumanEval (code generation), BFCL (function calling)
Four base models: Qwen3-1.7B-Base, Qwen3-4B-Base, SmolLM3-3B-Base, gemma-3-4b-pt
Each run: Single H100 GPU, 10-hour budget, fully autonomous, no human intervention; all agent roles running on Claude Opus 5
| Method | AIME | ArenaHard | BFCL | GPQA | GSM8K | HealthBench | HumanEval | Overall |
|---|---|---|---|---|---|---|---|---|
| AIBuildAI (ours) | 15.8 | 65.5 | 95.8 | 31.4 | 82.3 | 42.6 | 69.0 | 46.6 |
| Locus (Opus 5) | 9.4 | 66.3 | 94.3 | 33.1 | 82.9 | 44.3 | 66.6 | 45.6 |
| Claude Code (Fable 5) | 13.3 | 61.5 | 73.0 | 28.3 | 83.5 | 40.1 | 58.4 | 41.8 |
| Claude Code (Opus 5) | 9.2 | 53.4 | 1.5 | 31.4 | 80.3 | 38.7 | 59.8 | 35.0 |
| Official instruct (reference) | 29.2 | 70.2 | 85.0 | 36.2 | 87.0 | 43.3 | 71.5 | 51.1 |
Each cell is the average across four base models; bold = highest agent score
Analysis of three leading tasks:
BFCL (function calling, 95.8): The task is gated by a single output contract — the emitted call must parse and match the expected schema. Learn the contract, score high; miss it, score near zero. Meta search can precisely design a pipeline for this contract.
HumanEval (code generation, 69.0): Leads on all four base models. Code tasks have clear automated acceptance criteria (test pass rates), providing clean reward signals that reward iterative optimization.
AIME 2025 (competition math, 15.8): Math reasoning requires strict output formatting (integer answers), and the SFT + GRPO stacking strategy varies dramatically across small models. Meta search can adjust strategy based on intermediate results; fixed topologies can’t course-correct after early wrong decisions.
Core Technical Differences
Locus (external submission by Intology, also runs on Opus 5): 45.6 overall, leads on ArenaHard, GPQA, HealthBench. The current strongest single-agent external submission. AIBuildAI trails by less than 2 points on each of those three tasks.
Claude Code + Fable 5 (41.8): High variance on BFCL (73.0 ± 28.0 std dev), indicating inconsistent handling of function-calling format — Claude Code is a general coding agent, not designed around post-training domain knowledge.
Claude Code + Opus 5 (35.0): BFCL score of 1.5 — a near-complete formatting failure. The same underlying model (Opus 5) produces dramatically different results depending on the surrounding architecture.
Current Limitations
- Requires Linux x86_64 + systemd + cgroup-v2 — no macOS support
- Needs a single H100 per run (10-hour budget) — hardware costs are real
- All agents run on Claude Opus 5 — API costs stack on top of training costs
- Repository just opened (2026-09-16), 7 stars — production maturity is unknown
- PostTrainBench results are described as a “snapshot of a system still under construction”
The technical direction is clear: meta search + domain knowledge base + automated experimentation = more efficient post-training exploration than general-purpose agents. This aligns directly with the “human intent + agent execution” pattern that defines current AI engineering practice.
Open-source code and models are for learning purposes only — do not use directly in production work.
© 2026 Author: Mycelium Protocol. Licensed under CC BY 4.0 — free to share and adapt with attribution.
关于本站 · 免责声明
🍄 Mushroom Research Blog 是非营利、免费公开的个人科技观察博客与公众号 XStack18,不接受商业合作、不代表任何企业或机构立场,也不谋求商业利益。我们以个人视角客观中立地记录和分析 AI、Web3 等领域的最新模型发布与技术动态——不止转述新闻标题或二手信息,而是给出有独立思考的深入分析,希望帮更多人获得有价值的一手科技认知。
⚠️ 文中介绍的开源代码与模型,仅供学习交流与技术借鉴。它们大多仍处于早期阶段,有待进一步研究和验证,请勿直接用于工作或生产环境;如需采用,请先自行充分测试,并核实其许可证与安全性。
Open-source code and models featured here are shared for learning and reference only. Most are early-stage and still need further study and verification — please don't use them directly in your work or in production. Test them thoroughly and check their licenses and security first.
- 本站文章均为作者基于公开信息的个人研究与观点整理,不代表文中提及的任何公司、产品、模型的官方立场,未与其构成商业关联或合作关系。
- 科技行业信息更新极快,我们尽力保证内容准确、及时,但不对完整性、实时性做绝对保证,具体请以相关企业/项目官方公告为准。
- 文中引用的第三方商标、产品名称、图片、数据等版权归原权利人所有,我们会尽量注明来源;如你认为存在版权疑问或侵权,请通过下方邮箱联系我们,收到通知后会尽快核实处理(更正、加注来源或删除)。
- 文章内容仅为技术科普与个人观点,不构成投资、法律或其他专业建议,据此进行任何决策的后果需自行判断和承担。
📮 侵权 / 勘误 / 合作咨询:[email protected]
💬 评论与讨论
使用 GitHub 账号登录后发表评论