Skillget

genpark-kullback-leibler-divergence-drift-monitor-skill

mcp
Guvenlik Denetimi
Uyari
Health Uyari
  • No license — Repository has no license file
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 7 GitHub stars
Code Gecti
  • Code scan — Scanned 6 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

Forward and reverse KL divergence monitor detecting policy drift and reward hacking across distributions

README.md

genpark-kullback-leibler-divergence-drift-monitor-skill

Telemetry monitor calculating forward, reverse, and symmetric Kullback-Leibler (KL) divergence to prevent reward hacking.

Architecture

flowchart LR
    P["Active Policy pi_theta"] --> KL["Kullback-Leibler Engine"]
    Q["Reference Policy pi_ref"] --> KL
    KL --> Fwd["Forward KL(P || Q)"]
    KL --> Rev["Reverse KL(Q || P)"]
    Fwd --> Guard{KL > Threshold?}
    Guard -->|Yes| Alert[Trigger Drift Alert / Step Rollback]
    Guard -->|No| Safe[Policy Within Safe Bounds]

Features

  • Symmetric & Asymmetric Drift: Tracks both mode collapse and out-of-distribution tokens.
  • Zero Dependencies: 100% Python Standard Library.

Yorumlar (0)

Sonuc bulunamadi