Published January 7, 2026 | Version v1.0.0
Software Open

enfuse/ala-paper: ALA v1.0.0 - Initial Release

Authors/Creators

  • 1. @enfuse

Description

ALA: Asynchronous LLM Advisor

Bounded Logit Perturbation Channels for LLM-Guided Reinforcement Learning

Author

Cahlen Humphreys, Enfuse Labs

Abstract

A novel architecture that enables large language models to provide real-time strategic guidance to reinforcement learning agents via bounded logit perturbation channels with importance sampling correction for PPO training.

Key Innovations

  • Time-bounded bias expiration
  • Multi-advisor voting
  • Importance sampling correction for unbiased policy gradients

License

CC BY 4.0

Links

  • Website: https://mc.enfuse.ai
  • Paper PDF included in release assets

Files

enfuse/ala-paper-v1.0.0.zip

Files (413.5 kB)

Name Size Download all
md5:f6cba46600196fbd308e7cfbac0ab7f6
413.5 kB Preview Download

Additional details

Related works

Is supplement to
Software: https://github.com/enfuse/ala-paper/tree/v1.0.0 (URL)

Software

Repository URL
https://github.com/enfuse/ala-paper