MyTorch – Minimalist autograd in 450 lines of Python

https://github.com/obround/mytorch

93•iguana2000•1d ago

Comments

jjzkkj•1d ago

HmcKk

jerkstate•23h ago

Karpathy’s micrograd did it first (and better); start here: https://karpathy.ai/zero-to-hero.html

richard_chase•22h ago

Harsh.

whattheheckheck•21h ago

Why is it better

tfsh•21h ago

Because it's an acclaimed, often cited course by a preeminent AI Researcher (and founding member of OAI) rather than four undocumented python files.

nurettin•21h ago

Objective measures like branch depth, execution speed, memory use and correctness of the results be damned.

CamperBob2•20h ago

Karpathy's implementation is explicitly for teaching purposes. It's meant to be taken in alongside his videos, which are pretty awesome.

gregjw•20h ago

it being acclaimed is a poor measure of success, theres always room for improvement, how about some objective comparisons?

geremiiah•16h ago

Ironically the reason Karpathy's is better is because he livecoded it and I can be sure it's not some LLM vomit. Unfortunately, we are now indundated with newbies posting their projects/tutorials/guides in the hopes that doing so will catch the eye of a recuiter and land them a high paying AI job. That's not so bad in itself except for the fact that most of these people are completely clueless and posting AI slop.

iguana2000•15h ago

Haha, couldn't agree with you more. This, however, isn't AI slop. You can see in the commit history that this is from 3 years ago

forgotpwd16•16h ago

Cleaner, more straightforward, more compact code, and considered complete in its scope (i.e. implement backpropagation with a PyTorch-y API and train a neural network with it). MyTorch appears to be an author's self-experiment without concrete vision/plan. This is better for author but worse for outsiders/readers.

P.S. Course goes far beyond micrograd, to makemore (transfomers), minbpe (tokenization), and nanoGPT (LLM training/loading).

alkh•21h ago

Imho, we should let people experiment as much as they want. Having more examples is better than less. Still, thanks for the link for the course, this is a top-notch one

iguana2000•17h ago

Karpathy's material is excellent! This was a project I made for fun, and hopefully provides a different perspective on how this can look

khushiyant•15h ago

Better readme would be way to go

CamperBob2•10h ago

In iguana2000's defense, the code is highly self-documenting.

It arguably reads cleaner than Karpathy's in some respects, as he occasionally gets a little ahead of his students with his '1337 Python skillz.

brandonpelfrey•13h ago

Having written a slightly more involved version of this recently myself I think you did a great job of keeping this compact while still readable. This style of library requires some design for sure.

Supporting higher order derivatives was also something I considered, but it’s basically never needed in production models from what I’ve seen.

iguana2000•7h ago

Thanks! I agree about the style

Rapid Validation of Product Concepts with AI

Somebody Build This

Who's in charge of Venezuela and what happens next?

Show HN: CloudSlash – Find AWS waste and generate Terraform state rm commands

AGI Is Here

'Chinese Peptides' Are the Latest Biohacking Trend in the Tech World

They Said AI Would Replace You by Now

Americans Choosing Cremation at Historic Rates, NFDA Report Finds

Damn Vulnerable AI Bank – Practice AI Security

Show HN: A Android Color Detection Auto Clicker with no full-screen ads

Berlin power outages after left-wing anarchist attack on power cables

Don't Forget the WAL: How I Lost SQLite Data in Podman Containers

Wanderly AI Travel App Waitlist

During Helene, I Just Wanted a Plain Text Website

Agent Orchestration Is Not the Future

What is Agent context engine

Tempest Future Fighter Aims for "Extreme Range," Twice F-35 Payload

Politics and the English Language – George Orwell [Essay]

Show HN: Vho – AST-based analysis for better AI refactoring of large codebases

vLLM: An Efficient Inference Engine for Large Language Models

Linuxulator on FreeBSD Feels Like Magic

Ask HN: What app features actually help vocabulary stick long-term?

Ask HN: Is there a better alternative to email?

AI Safety ArXiv Scraper

Translating Cave Story into Classical Latin with Gemini

Show HN: I Made a Gamma Clone with 1 Prompt

Cool project, will you actually maintain it?

The State of LLMs 2025: Progress, Problems, and Predictions

The Intelligent Universe: AI, ET, and the Emerging Mind of the Cosmos

Physics of Language Models: How to Build Versatile Pretrain Playgrounds [video]