Unverified60% confidenceFactTime unknown

字节跳动面试高频考察RLHF的GenericGPT → SFT → RL三阶段训练流程

1

Sources

60%

Confidence

Medium-term (~90 days)

Relevance

6/1/2026

First Seen

Valid until: 8/30/2026

Sources

字节腾讯阿里AI Agent面试重点对比：考法差异与备考攻略

bilibili海云日记

Related Entities

字节跳动company RLHFconcept

Related Claims

VerifiedSFT-奖励模型-RLHF三阶段对齐训练范式最早由OpenAI在InstructGPT论文中系统提出75% similar Verified从预训练检查点到可商用产品需经历监督式指令微调（SFT）和基于人类反馈的强化学习（RLHF）两个后训练阶段74% similar Verified大模型训练经历三个核心阶段：预训练、有监督微调（SFT）和偏好对齐72% similar Unverified后训练通常涵盖监督微调（SFT）、强化学习对齐（RLHF/RLAIF）以及领域适配微调，视觉推理模型还涉及视觉指令微调71% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/20148

API

curl https://kongchang.com/api/v1/knowledge/claims/20148

MCP

get_claim(id=20148)