Carlos Villuendas Zambrana carlosvillu

LLM Wiki

A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

The core idea

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

When loading the LoRA params (that were obtained on a quantized base model) and merging them into the base model, it is recommended to first dequantize the base model, merge the LoRA params into it, and then quantize the model again. This is because merging into 4bit quantized models can lead to some rounding errors. Below, we provide an end-to-end example:

First, load the original model and merge the LoRA params into it:

from diffusers import FluxPipeline 
import torch 

ckpt_id = "black-forest-labs/FLUX.1-dev"
pipeline = FluxPipeline.from_pretrained(

Articles

NOTICE: I moved most of this content that I no longer maintain within this gist over to its own repo here https://github.com/radum/performance-testing-framework

Performance API

// https://github.com/WICG/paint-timing
performance.getEntries();

You can install Node Nightlies/RCs via nvm using NVM_NODEJS_ORG_MIRROR environment variable.

Install latest Node RC

NVM_NODEJS_ORG_MIRROR=https://nodejs.org/download/rc/ nvm i node

Install latest Node.js Nightly

NVM_NODEJS_ORG_MIRROR=https://nodejs.org/download/nightly/ nvm i node

	from diffusers import StableDiffusion3Pipeline
	from transformers import T5EncoderModel
	import torch
	import time
	import gc

	def flush():
	gc.collect()
	torch.cuda.empty_cache()

	- Be highly organized.
	- Suggest proactive solutions and anticipate my needs.
	- Treat me as an expert in all subject matter.
	- Be accurate and thorough; mistakes erode my trust.
	- Provide detailed explanations; I appreciate lots of detail.
	- Value good arguments over authorities; the source is irrelevant.
	- Consider new technologies and contrarian ideas.
	- High levels of speculation or prediction are fine; just flag it.
	- Recommend only the highest-quality, meticulously designed products.
	- Recommend products from all over the world; location is irrelevant.

	"use strict";

	[foo,bar] = TNG(foo,bar);
	// NOTE: intentionally not TNG(..) wrapping useBaz(), so that it's
	// basically like a "custom hook" that can be called only from other
	// TNG-wrapped functions

	function foo(origX,origY) {
	var [x,setX] = useState(origX);
	var [y,setY] = useState(origY);

	var paths = [
	["Account"],
	["Account", "Payment Methods"],
	["Account", "Payment Methods", "Credit Card"],
	["Account", "Payment Methods", "Paypal"],
	["Account", "Emails"],
	["Account", "Emails", "Main Email"],
	["Account", "Emails", "Backup Email"],
	["Account", "Devices"],
	["Account", "Devices", "Google Pixel"],

	import React from "react";
	import ReactDOM from "react-dom";
	import { AppContainer } from "react-hot-loader";

	import App from "./App";

	const render = (hydrate = false) => {
	const container = document.querySelector("#app");
	const element = (
	<AppContainer>

	// routes.js
	const routes = [
	{
	path: '/',
	component: Home,
	exact: true
	},
	{
	path: '/gists',
	component: Gists