gitwtfhub

wtf is lecun1989-repro?

karpathy/lecun1989-repro — explained in plain English

Analysis updated 2026-07-18 · repo last pushed 2024-02-03

★ 765Jupyter NotebookAudience · researcherComplexity · 2/5DormantSetup · easy

TL;DR

A reproduction of Yann LeCun's landmark 1989 zip-code recognition experiment, retraining the original neural network architecture on modern hardware in 90 seconds instead of three days.

Mindmap

mindmap
  root((lecun1989-repro))
    What it does
      Recreates 1989 experiment
      Shrinks MNIST dataset
      Trains original architecture
    Tech Stack
      Python
      Jupyter Notebook
    Use Cases
      Study neural network history
      Benchmark hardware speedup
      Learn from famous code
    Audience
      ML researchers
      Students
      History enthusiasts

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

Why would anyone build with this?

REASON 1

Study how early backpropagation-trained neural networks were designed and connected.

REASON 2

See a concrete benchmark of how much faster modern hardware trains the same 1989 algorithm.

REASON 3

Learn how researchers fill in missing implementation details when reproducing old papers.

What's in the stack?

PythonJupyter Notebook

How it stacks up

karpathy/lecun1989-reprollsourcell/how-to-predict-stock-prices-easily-demokarpathy/deep-vector-quantization
Stars765771647
LanguageJupyter NotebookJupyter NotebookJupyter Notebook
Last pushed2024-02-032022-06-232021-11-20
MaintenanceDormantDormantDormant
Setup difficultyeasymoderatehard
Complexity2/52/54/5
Audienceresearchervibe coderresearcher

Figures from each repo's GitHub metadata at analysis time.

How do you spin it up?

Difficulty · easy Time to first run · 5min

Runs in about 90 seconds on a modern laptop, versus three days on 1989-era hardware.

Yoink these prompts

Prompt 1
Explain how this repo shrinks the modern MNIST dataset to match the original 1989 zip-code dataset.
Prompt 2
Walk me through the architectural quirks in this reproduction that differ from modern neural networks.
Prompt 3
Compare the error rates this reproduction achieves to the original 1989 paper's results.
Prompt 4
Help me run this notebook locally to reproduce the 1989 handwriting recognition experiment myself.

Frequently asked questions

wtf is lecun1989-repro?

A reproduction of Yann LeCun's landmark 1989 zip-code recognition experiment, retraining the original neural network architecture on modern hardware in 90 seconds instead of three days.

What language is lecun1989-repro written in?

Mainly Jupyter Notebook. The stack also includes Python, Jupyter Notebook.

Is lecun1989-repro actively maintained?

Dormant — no commits in 2+ years (last push 2024-02-03).

How hard is lecun1989-repro to set up?

Setup difficulty is rated easy, with roughly 5min to a first successful run.

Who is lecun1989-repro for?

Mainly researcher.

View the repo → Decode another repo

This repo across BitVibe Labs

Don't trust strangers blindly. Verify against the repo.

Other repos in this lane