Warp-net/warpnet

Add LLM-Based or SLM Swarm-based Moderation Node for WarpNet

Offen

#3 geöffnet am 20.05.2025

 (3 Kommentare) (0 Reaktionen) (0 zugewiesene Personen)Go (12 Forks)auto 404
AIenhancementhelp wantedpartially-done

Repository-Metriken

Stars
 (137 Sterne)
PR-Merge-Metriken
 (PR-Metriken ausstehend)

Beschreibung

In order to enforce basic human rights and safety across the WarpNet distributed social network, we need to introduce a dedicated moderation node powered by a local LLM (Large Language Model). This node should be capable of automatically evaluating content and rejecting harmful material such as:

  • Child sexual abuse material (CSAM)
  • Violent or gore content
  • Illegal drug or weapon sales
  • Calls for terrorism or extremist violence etc.

Given the decentralized nature of WarpNet, content moderation must happen locally, without reliance on centralized infrastructure or cloud APIs. A locally-run LLM enables us to evaluate and flag content as "hidden".

Possible solutions:

  • Integrate a local LLM (such as Mistral or LLaMA) into a dedicated node.
  • Use llama.cpp or onnxruntime for inference, eliminating the need for Docker or external APIs.
  • Define a simple moderation API (/moderate) to evaluate content before storage or broadcast.
  • Optionally distribute moderation tags to peer nodes to assist in propagation control.

Good to have:

  • Allow plug-in moderation models (e.g., community-trained LoRA adapters)
  • Support image and video moderation via CLIP or ONNX-based classifiers
  • Log moderation decisions with transparency (e.g., why a post was rejected)

Ways to integrate with the network:

  • Pubsub
  • Requesting list of tweets worker
  • (Your solution)

Please help!

Contributor Guide