karpathy

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.

BIGBALLON

#!/bin/bash
#
# script to extract ImageNet dataset
# ILSVRC2012_img_train.tar (about 138 GB)
# ILSVRC2012_img_val.tar (about 6.3 GB)
# make sure ILSVRC2012_img_train.tar & ILSVRC2012_img_val.tar in your current directory
#
#  https://github.com/facebook/fb.resnet.torch/blob/master/INSTALL.md
# 
#  train/

zhreshold

name: "ResNet-18"
input: "data"
input_dim: 1
input_dim: 3
input_dim: 224
input_dim: 224

layer {
	bottom: "data"
	top: "conv1"

gauravkaila

#!/bin/bash
# add the GPG key for the official Docker repository to the system
curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo apt-key add -

# add the Docker repository to APT sources
sudo add-apt-repository "deb [arch=amd64] https://download.docker.com/linux/ubuntu $(lsb_release -cs) stable"

# update the package database with the Docker packages from the newly added repo
sudo apt-get update

MInner

import datetime
import linecache
import os

import pynvml3
import torch

print_tensor_sizes = True
last_tensor_sizes = set()
gpu_profile_fn = f'{datetime.datetime.now():%d-%b-%y-%H:%M:%S}-gpu_mem_prof.txt'

jychstar

import argparse
import sys

from tensorflow.examples.tutorials.mnist import input_data
from time import time
t0 = time()

import tensorflow as tf

tf.summary.FileWriterCache.clear()

max-mapper

Brainiarc7

Building Tensorflow from source on Ubuntu 16.04LTS for maximum performance:

TensorFlow is now distributed under an Apache v2 open source license on GitHub.

On Ubuntu 16.04LTS+:

Step 1. Install NVIDIA CUDA:

To use TensorFlow with NVIDIA GPUs, the first step is to install the CUDA Toolkit as shown:

batzner

import sys, getopt

import tensorflow as tf

usage_str = 'python tensorflow_rename_variables.py --checkpoint_dir=path/to/dir/ ' \
            '--replace_from=substr --replace_to=substr --add_prefix=abc --dry_run'


def rename(checkpoint_dir, replace_from, replace_to, add_prefix, dry_run):
    checkpoint = tf.train.get_checkpoint_state(checkpoint_dir)

wllhf

""" 
Python implementation of the color map function for the PASCAL VOC data set. 
Official Matlab version can be found in the PASCAL VOC devkit 
http://host.robots.ox.ac.uk/pascal/VOC/voc2012/index.html#devkit
"""
import numpy as np
from skimage.io import imshow
import matplotlib.pyplot as plt

def color_map(N=256, normalized=False):