feat(wiki): ingest misc sources batch (Daily notes/Others/Skills/Vibe Coding/微信公众号/跨境电商)
This commit is contained in:
25
wiki/concepts/验证器模式.md
Normal file
25
wiki/concepts/验证器模式.md
Normal file
@@ -0,0 +1,25 @@
|
||||
---
|
||||
title: "验证器模式"
|
||||
type: concept
|
||||
tags: []
|
||||
---
|
||||
|
||||
## Definition
|
||||
多智能体系统中 Validator(验证器)检查 Worker 输出质量,不合格则退回的机制。验证器可以是确定性代码(单元测试、JSON Schema 验证)或 LLM 本身。
|
||||
|
||||
## Validation Methods
|
||||
- **确定性代码验证**:单元测试、JSON Schema 验证、正则表达式匹配
|
||||
- **LLM 验证**:单独训练或提示的验证模型判断输出质量
|
||||
- **双模式**:可单独验证每个 Worker 输出,或在汇总所有结果后整体验证
|
||||
|
||||
## Placement in Multi-Agent Patterns
|
||||
- [[层级结构]](Hierarchy):Validator 是关键角色,位于 Planner → Worker → Validator 链路的末端
|
||||
- [[淘汰制]](Knock-out):Validator 决定哪些 Agent 被淘汰
|
||||
- [[对抗辩论]](Adversarial Debate):Watchdog(确定性代码)打破辩论死循环
|
||||
|
||||
## Best Practice
|
||||
- Validator 最好使用与 Generator/Worker 不同的模型(提高质量和客观性)
|
||||
- 验证器可以在同一 LLM 会话中与规划器协作(PLAN → VALIDATION loop)
|
||||
|
||||
## Sources
|
||||
- [[Multi-Agent-System-Reliability.md]]
|
||||
Reference in New Issue
Block a user