🔥我靠！终于有大公司把AI模型后训练的“魔法”拆成可复现配方了！
This is a dev post classified by AGK Intelligence as AI dev tools (news), kept by the Dev Radar because it carries real work, not commentary.
🔥我靠！终于有大公司把AI模型后训练的“魔法”拆成可复现配方了！ Amazon团队直接公开从GLM-4.5-Air-Base出发的完整8阶段串行后训练pipeline，数据来源、奖励设计、筛选方法、训练配置、阶段结果全部写清楚，且大量用公开数据、无新增人工标注、无内部蒸馏教师！ 他们发现： 1️⃣串行8阶段流水线（每一步继承上一步参数）： SFT → Reasoning RL → Coding RL → Instruction-Following RL → General Agent → Coding Agent → Search Agent → RLHF 从基础能力到高级Agent，从硬可验证奖励逐步过渡到软judge奖励。 2️⃣不造新数据也能大幅提升：大量复用已公开数据集，附录详细标注来源，证明高质量公开数据+合理筛选就能打。 3️⃣核心原则： ▫️多样高质量SFT先打牢能力地板 ▫️难度过滤让RL prompt保持在有效学习区间 ▫️奖励可靠性决定阶段顺序（先硬规则/执行验证，再软LLM judge） ▫️工程与基础设施本身就是配方的一部分 4️⃣效果炸裂：Rufus-Air在多个基准上大幅超越智谱官方后训练的GLM-4.5-Air，与同规模开源模型竞争激烈。 后训练不再是黑箱！这套recipe直接拉高了开源可复现的天花板。 论文链接：https://arxi
Posted by SuSu_酥酥👅 (26.8k followers) 48 min ago · 21 likes · 1.2k views · view the original post on X. Kept by the Dev Radar as AI dev tools. Tools mentioned: arXiv.org.
More dev work like this
- Attention all Claude Code users ! — @ItsmeAjayKV
- The first results for Terminal Bench 2.1, all 89 tasks, 3 iterations. Our ~1,300 lines… — @PawelHuryn
- 😱惊了啊！100万行神级DeepSeek Harness的框架，被他用1000行TypeScript彻底扒光！ — @NFT_Chen
- 想学 AI Agent，搜出来的教程东一篇西一篇，有的上来就讲多 Agent 协作，连 Token 是什么都还没弄明白。 — @GitHub_Daily
- Build an AI workflow visually, then publish it as a chat app. — @DanKornas
- I found a setup that saves you $300 a month on AI tools and you don't lose much — @VaibhavSisinty
- Hermes supremacy tweet #9138 — @Teknium
- OpenAI 前 CTO Mira Murati 创办的 Thinking Machines 发布新模型 Inkling 2 Small。Artificial Analysis… — @0xLogicrw
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.9k posts from 5k X accounts over the last 21 days, 9.5k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-06 08:46 UTC. Full method.