ruvnet

Building poison-pill: A Rust-WASM Document Sanitization Middleware

Introduction

Document poisoning attacks represent a critical and unsolved vulnerability in LLM applications. Research shows just 250 malicious documents can backdoor LLMs of any size, while 5 poisoned documents can compromise RAG systems with millions of entries. This implementation plan provides a complete roadmap for building “poison-pill” - a high-performance, Rust-based WASM middleware that sanitizes documents before they reach LLMs, distributed via npm and executable through npx.

---

1. DOCUMENT POISONING ATTACKS: The Threat Landscape

lfhbento

// ==UserScript==
// @name         Kindle Download
// @namespace    http://tampermonkey.net/
// @version      2025-02-20
// @description  Download all your kindle books
// @author       You
// @match        https://www.amazon.com/hz/mycd/digital-console/contentlist/booksPurchases/*
// @icon         https://www.google.com/s2/favicons?sz=64&domain=amazon.com
// @grant        none
// ==/UserScript==

btbytes

PostgreSQL is Enough

• Simplify: move code into database functions
• Just Use Postgres for Everything
• PostgreSQL is the worlds’ best database
• Postgres is eating the database world
• Hacker News discussion

Background and Cron Jobs

Artefact2

Which GGUF is right for me? (Opinionated)

Good question! I am collecting human data on how quantization affects outputs. See here for more information: ggml-org/llama.cpp#5962

In the meantime, use the largest that fully fits in your GPU. If you can comfortably fit Q4_K_S, try using a model with more parameters.

llama.cpp feature matrix

See the wiki upstream: https://github.com/ggerganov/llama.cpp/wiki/Feature-matrix

pixelass

{
  "LoRA_type": "Standard",
  "LyCORIS_preset": "full",
  "adaptive_noise_scale": 0,
  "additional_parameters": "",
  "block_alphas": "",
  "block_dims": "",
  "block_lr_zero_threshold": "",
  "bucket_no_upscale": true,
  "bucket_reso_steps": 64,

btbytes

mberman84

git clone https://github.com/cpacker/MemGPT.git
conda create -n memgpt python=3.10
conda activate memgpt
cd memgpt
pip install -r requirements.txt
export OPENAI_API_KEY="YOUR_API_KEY"
# download .txt docs from: https://huggingface.co/datasets/MemGPT/example-sec-filings/tree/main
# place in "memgpt/personas/examples/preload_archival/*.txt"
python3 main.py --archival_storage_files_compute_embeddings="memgpt/personas/examples/preload_archival/*.txt" --persona=memgpt_doc --human=basic

btbytes

-> <-

The Novice's LLM Training Guide

->Written by Alpin<- ->Inspired by /hdg/'s LoRA train rentry<- !!!warning This guide is being slowly updated. We've already moved to the axolotl trainer.

---

[TOC2]

mberman84

# this tutorial assumes conda and git are both installed on your computer

conda create -n tg python=3.10.9
conda activate tg
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117
git clone https://github.com/oobabooga/text-generation-webui.git
cd text-generation-webui
pip install -r requirements.txt

# GPU only:

vizsumit

{
  "LoRA_type": "Standard",
  "adaptive_noise_scale": 0,
  "additional_parameters": "",
  "block_alphas": "",
  "block_dims": "",
  "block_lr_zero_threshold": "",
  "bucket_no_upscale": true,
  "bucket_reso_steps": 64,
  "cache_latents": true,