Skip to content

docs: 新增基础模型训练决策专题 T01 - #38

Merged
study8677 merged 1 commit into
mainfrom
codex/llm-training-decision-t01
Sep 7, 2026
Merged

study8677 merged 1 commit into
mainfrom
codex/llm-training-decision-t01

Conversation

@study8677

Copy link
Copy Markdown
Owner

这次改了什么

  • 新增独立专题首章《训练方案不是算法排行榜:先把决策分层》
  • 从 Policy、Rollout、Reward、Critic、Advantage 等基础角色讲起,再解释 SFT、DPO、PPO、GRPO
  • 用“目标行为 → 信号来源 → 数据与轨迹 → 奖励粒度 → 信用分配 → 策略更新 → 系统拓扑 → 评测门禁”八层框架组织训练决策
  • 通过 GLM-5 到 GLM-5.2 的公开案例,展示任务与轨迹约束如何推动 group-wise optimization 向 critic-based PPO 调整
  • 补充决策矩阵、ADR 模板、失败模式、自测题和术语表

为什么这样写

本章不讨论普通团队是否有资格从头预训练,也不把算法按“先进程度”排序。它只解决一个问题:团队已经进入基础模型生产之后,怎样让任务结构、学习信号、信用分配与训练系统共同决定方案。

本次边界

  • 只增加一个可独立阅读的专题章节
  • 不调整现有 1—40 章编号
  • 不修改站点导航和 README
  • 不在本次 PR 中展开后续训练工程章节

验证

  • npm run docs:build
  • git diff --check origin/main...HEAD
  • 本地 Markdown 相对链接存在性检查

VitePress 构建通过;仅保留仓库现有的 chunk 大小提示。

@study8677
study8677 marked this pull request as ready for review September 7, 2026 12:30
@study8677
study8677 merged commit 66c73ef into main Sep 7, 2026
1 check passed
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Sep 7, 2026 •

Copy link
Copy Markdown

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-09-07T12:31:40.509540Z 0368919 Draft marked ready
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant