dubzdubz

dubzdubz / microgpt-ts

Public

A complete GPT built from scratch in TypeScript with zero dependencies

microgpt-ts.vercel.app gpt learning llm microgpt typescript
20
4
100% credibility
Found Feb 25, 2026 at 18 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

A browser-based playground to train a tiny character-level language model from scratch on text datasets and generate new samples.

How It Works

1
🌐 Discover the playground

You land on a simple webpage and spot the inviting playground button promising to train your own tiny AI right in the browser.

2
📝 Pick your words

Choose from fun lists like baby names or Pokémon, or paste your own words, one per line, to teach the AI your style.

3
⚙️ Tweak the settings

Adjust the brain size or training length if you like, or just hit go with the easy defaults.

4
🚀 Watch it learn

Click train and see the graph drop as loss shrinks, with live new words popping up to show it's getting smarter.

5
✨ Generate creations

Switch to generate, set randomness, add a starting word, and create batches of fresh inventions.

6
Explore step by step
🎲
Quick batch

Generate many at once for instant fun.

🔍
Token explorer

Peek inside the AI's mind, one prediction at a time.

🎉 Your AI masterpieces

Delight in endless new names, titles, or words that feel perfectly yours, all created in seconds on your screen.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 18 to 20 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 microgpt-ts?

microgpt-ts delivers a full GPT for training and inference on small text datasets, running entirely in the browser with zero runtime dependencies. Load presets like baby names or Pokémon, tweak architecture like embedding size or layers, and train via a live playground that shows loss curves, sample generations, and token-by-token probabilities. Developers get instant experimentation on character-level language modeling, no setup beyond opening a page.

Why is it gaining traction?

It stands out by ditching heavy ML frameworks for pure TypeScript math, letting anyone train a transformer in seconds without servers or installs—ideal for a complete chat gpt tutorial. The step-by-step browser UI hooks users with real-time visuals like live samples during training and interactive generation explorers, turning abstract concepts into tangible demos. Early adopters praise the educational progression from bigrams to full multi-head attention.

Who should use this?

ML educators building curriculum complete gpt modules, frontend devs prototyping text gen UIs, or hobbyists training on custom lists like company names from S&P 500 data. Suited for quick experiments on short sequences, not production-scale models—think download complete github repo and toy with pokemon names in under a minute.

Verdict

Grab it for hands-on transformer learning; the playground nails browser-based training better than most demos. At 16 stars and 1.0% credibility, it's raw and lightly tested—fine for education, but scale up elsewhere for real apps.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.