frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Show HN: MicroGPT-C – C99 GPT for Edge Training and Tiny Model Pipelines

https://github.com/enjector/microgpt-c
1•Ajay__soni•1h ago
TL;DR: Pure C99 GPT-2 engine, zero dependencies. Sub-1M-parameter “organelles” coordinate via a Planner-Worker-Judge pipeline and beat random opponents up to 91 % on logic games, all on CPU.

I’m a C/C++ architect focused on low-latency systems. Last year, I tried building agentic pipelines with SLMs/LLMs and hit the usual wall: latency and orchestration overhead killed real-time edge use cases.

Initial research video: https://www.youtube.com/watch?v=q-rs9VZ1-0I

So I asked: how far can you push specialised logic at <1M parameters with nothing but local CPU?

MicroGPT-C is a from-scratch C99 port of Karpathy’s microgpt.py (https://gist.github.com/karpathy/8627fe009c40f57531cb1836010...). Zero deps, single-header, localised KV cache. Speed was never the goal (Andrej’s nanoGPT and llm.c already showed what’s possible). The real experiment was orchestration.

Organelle Pipeline Architecture (OPA): Agile-style Planners, Workers and Judges talking through tiny structured strings (board=XO_|valid=1,3) parsed by a safety-gated stack VM (3.7–5.8M ops/s). A 64K model needed 181 manual interventions; a 460K model trained on those traces internalised everything and needed zero.

Beyond the research: fully auditable AI, great for education (~3k lines of readable C), rapid prototyping, and embedded.

Personal itch: fraud/risk engines. I want agents that hunt “unknown-unknowns” in a sandbox where every decision is inspectable.

Happy to talk implementation, the 97 tests, 22 benchmarks, or anything else. FAQ: https://github.com/enjector/microgpt-c/blob/main/FAQ.md

Quick try (macOS/Linux/Windows): git clone https://github.com/enjector/microgpt-c && cd microgpt-c mkdir build && cd build && cmake .. -DCMAKE_BUILD_TYPE=Release && cmake --build . ./connect4_demo # 460K params, ~21 min train, 88% win rate vs random

Performance (Apple M2 Max): - 4.2K params names: 685k tok/s train, 110k tok/s infer - 841K Shakespeare char: 28k / 16k tok/s - 510K Shakespeare word: 12.5k / 40k tok/s

Full leaderboard (11 games), market-regime experiment (57% holdout = 2.8× baseline), and the book PDF: https://github.com/enjector/microgpt-c/blob/main/docs/book/M... GitHub: https://github.com/enjector/microgpt-c

The battle of the squares: When domestic workers protest

https://toolong.co/s/lz5q172x
1•a_void_sky•16s ago•0 comments

Ex-UK ambassador to U.S. Peter Mandelson arrested in London

https://www.cnbc.com/2026/02/23/epstein-files-peter-mandelson-london-uk.html
1•Betelbuddy•34s ago•0 comments

The Statistical Signature of LLMs

https://arxiv.org/abs/2602.18152
1•bikenaga•44s ago•0 comments

Exploiting Almost Every Antivirus Software (2020)

https://rack911labs.ca/research/exploiting-almost-every-antivirus-software/
1•LorenDB•58s ago•0 comments

From Select to Advanced SQL: JOINs, CTEs, and More

https://jsdev.space/sql-complete-guide/
1•javatuts•1m ago•0 comments

Show HN: Find Local Business Leads

https://leadsbylocation.com
1•corefiredrill•2m ago•0 comments

RWKV-7 beats Llama 3.2 with 3x fewer training tokens and formally exceeds TC^0

https://ai.gopubby.com/rwkv-7-beats-llama-3-2-rnn-constant-memory-46064bbf1f64
1•Aedelon•3m ago•1 comments

Show HN: BaitAI – Conversational AI interviews with realtime interview artifacts

https://baitai.club
1•Rupak182•3m ago•0 comments

The Laws of Software

https://www.laws-of-software.com/
1•itzlambda•4m ago•0 comments

Ask HN: How are you testing LLM integrations in CI?

1•akarshc•4m ago•0 comments

The Rise of Eyes Began with Just One

https://www.nytimes.com/2026/02/23/science/evolution-vertebrate-eye.html
2•marojejian•4m ago•1 comments

AOL (Sign On – Dial Up) [video]

https://www.youtube.com/watch?v=D1UY7eDRXrs
1•surprisetalk•5m ago•0 comments

Wilks' Tolerance Intervals

https://entropicthoughts.com/wilks-tolerance-intervals
1•surprisetalk•5m ago•0 comments

Unconventional strategies for practicing Spanish (2019)

https://devon.postach.io/post/unconventional-strategies-for-practicing-spanish
1•surprisetalk•5m ago•0 comments

Everything is Amazing, But Nothing is Ours (2019)

https://devon.postach.io/post/everything-is-amazing-but-nothing-is-ours
1•surprisetalk•5m ago•0 comments

Show HN: Brag Frog – The brag doc that builds itself from GitHub, Jira and Co

https://brag-frog.org/
1•recvonline•6m ago•0 comments

The best Part about receiving your miracle

https://in-due-time.com/when-your-receive-your-miracle/
1•marysminefnuf•7m ago•0 comments

Why SLMs are eating the world (and how to use them)

https://project-1960fbd1.doanything.app
1•aibuildersdig•7m ago•1 comments

Solubility in olive oil predicts anasthetic potency of a compound

https://en.wikipedia.org/wiki/Theories_of_general_anaesthetic_action
1•rdgthree•8m ago•0 comments

Snake Charmer – a modern way to view Python docs

https://snake-charmer.work/3.14/tutorial/appetite
1•spikepuppet•8m ago•1 comments

Inkplate13 Spectra - Color 13" eink board

https://www.crowdsupply.com/soldered/inkplate-13spectra
2•ashenke•8m ago•0 comments

Naiply – Real-time multiplayer Spanish card games built with Phoenix LiveView

https://naiply.com/
1•mendrugory•8m ago•1 comments

Hey I almost got scammed by Google

https://statmodeling.stat.columbia.edu/2026/02/21/hey-i-almost-got-scammed-by-google/
1•sega_sai•8m ago•0 comments

Show HN: Tastebuds AI – recipe generator with USDA-backed nutrition data

https://www.tastebudsai.com
1•marcosinghof•9m ago•1 comments

C64 Graphics Explorer

https://c64graphicsexplorer.com/
1•haunter•9m ago•0 comments

Why the EU's AI Act is about to become enterprises' biggest compliance challenge

https://techpinions.com/why-the-eus-ai-act-is-about-to-become-every-enterprises-biggest-complianc...
3•swolpers•10m ago•0 comments

Writing code is cheap now

https://simonwillison.net/guides/agentic-engineering-patterns/code-is-cheap/
4•swolpers•11m ago•0 comments

What is digital film mastering?

http://www.egpetersen.com/more/morestuff/dimstrg.htm
1•longislandguido•12m ago•0 comments

Show HN: Next.js Bot Prerender Middleware (Page Replica and Vercel)

https://github.com/jerjoh75/nextjs-page-replica-middleware
1•nirvanist•13m ago•0 comments

ASML unveils EUV light source advance that could yield 50% more chips by 2030

https://www.reuters.com/world/china/asml-unveils-euv-light-source-advance-that-could-yield-50-mor...
3•pieterr•13m ago•0 comments