jayyvk

karpathy's microgpt, ported to JavaScript and running entirely in your browser.

10
5
100% credibility
Found Mar 04, 2026 at 10 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
JavaScript
AI Summary

A browser-based interactive app for training a tiny language model on custom text lists like names or words and generating similar outputs, running completely on the user's device.

How It Works

1
🌐 Discover Train My Own GPT

You find and open a playful web page that lets you build and train a tiny AI brain right in your web browser, no setup needed.

2
📝 Pick your learning material

Choose from fun built-in lists like baby names, dinosaur types, or startup ideas, or simply drag and drop your own text file with words or names, one per line.

3
🔧 Shape your AI's mind

Use easy sliders to adjust the size of the brain, number of thinking layers, memory span, and training time to fit what you want.

4
🚀 Hit train and watch the magic

Click the start button to launch training, seeing live charts of progress as random numbers turn into a smart pattern learner before your eyes.

5
📊 Track the learning journey

Follow colorful graphs of improving scores, step speeds, and samples popping up, feeling the thrill as it masters your data.

Create new wonders

Type a starting word or letter, tweak creativity, and generate fresh names or ideas mimicking your data, all private on your device.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 10 to 10 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is trainmyowngpt?

Trainmyowngpt ports Karpathy's microgpt from his GitHub repo to JavaScript, letting you train a tiny GPT model entirely in your browser on custom data like startup names or baby names. Load built-in datasets or upload a .txt file, tweak architecture params like layers and context length via sliders, hit train, and watch loss drop in real-time charts. Generate text samples instantly after training—no servers, no data leaves your device.

Why is it gaining traction?

It stands out by running Karpathy's exact GPT math (inspired by his github.com/karpathy/nanogpt and llm-council) on CPU via Web Workers, with a slick draggable node UI visualizing dataset, tokenizer, config, training, metrics, and generation flows. Developers dig the privacy and speed for quick experiments, unlike cloud tools that leak data or need Python setups. The visual feedback hooks you, showing random weights turn into coherent outputs in seconds.

Who should use this?

ML educators demoing transformers to JS devs, frontend engineers building LLM toys without backends, or indie hackers training name generators on YC-style lists. Perfect for browser-bound workshops on Karpathy's github llm or makemore concepts, or rapid prototyping custom vocab models like dinosaur species.

Verdict

Fun, faithful browser port at ~4k params—great for learning Karpathy's techniques without setup. With 10 stars and 1.0% credibility score, it's early and lacks polish like tests or exports, but fork it for demos today.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.