Files
nexus/wiki/concepts/AirGappedSLMFixGeneration.md
2026-05-03 05:42:12 +08:00

38 lines
1.3 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
---
title: "Air-Gapped SLM Fix Generation"
type: concept
tags: []
last_updated: 2026-05-01
---
## Definition
在完全离线(气隙)的环境中,通过本地 Small Language ModelsSLM如 Ollama 运行的 Phi-3/Llama-3/Mistral生成确定性修复逻辑Python lambda的方法论。
## Core Principle
**AI generates logic — never touches data directly.**
SLM 输出的是一个转换函数lambda由系统执行而非 AI 直接修改数据。这样保证了可审计、可回滚、可解释的数据变更。
## Workflow
1. SLM 接收聚类样本 + 列名
2. SLM 输出严格格式化的 JSON含 transformation、confidence_score、reasoning、pattern_type
3. Lambda Safety Gate 验证(必须以 `lambda` 开头,不含 `import/exec/eval/os/subprocess`
4. 验证通过后向量化执行于整个聚类
5. 低于 0.75 置信度的自动进入人工隔离队列
## Safety Guarantees
- **Zero PII Egress**: 所有处理完全本地,无网络出口
- **Deterministic Output**: SLM 输出确定性 lambda不做创意性文本生成
- **Safety Gate**: 任何包含危险关键词的 lambda 立即被拒绝并路由至隔离区
- **Audit Trail**: 每条数据变更记录完整上下文
## Related
- [[Semantic Anomaly Compression]]
- [[Lambda Safety Gate]]
- [[AI Generates Logic Not Data]]