First screen, captured 1 day ago. Open the video.

I'm making LLMs work on small and old devices

What HN thinks

Will it run Doom has been replaced by will it run Qwen, and someone has already volunteered a vape with a screen.

Fun hack, suggest harder devices
Many cheered the project and pitched cursed targets like screen vapes, Playdates, PS2 dev kits, Speak and Spells and ESP32 boards.
What are tiny models for?
Several asked about practical use. Answers pointed to narrow fine-tuned tasks, speech, log parsing and messy data in regulated on-prem settings, though the author admits it is mostly novelty for now.
It's just Claude doing it
Skeptics said agents can one-shot inference engines now and called it slop. The author says the real work is steering a large brute-force search over kernels to get 16B models running.

Consensus: Most agree tiny models are getting more capable fast and that squeezing them onto old hardware is a fun engineering challenge, even if they aren't broadly useful yet.

Recommended in the comments

ps2-llm ↗ (tool): Offered as proof that the PS2 challenge has already been done.
iiama2.c ↗ (tool): A similar retro-hardware LLM effort.
esp32-ai ↗ (tool): Shows a technique for running models in kilobytes of RAM.
go-pherence ↗ (tool): Its author is pursuing the same goal and offered to swap notes.
Gemma (library): Named as a tiny model actually built for useful work, unlike scaled-down Qwen, and reported running on a GTX 1660.
MiniCPM (library): Suggested as better designed for tiny-scale work than Qwen.

Summarised by Claude from the comment thread 21 hours ago.

Points: 457
Comments: 173
Best rank: #4
Posted: 2026-11-12 by mrashad74
Category: AI & ML

llm-inference edge-ai llama-cpp quantization mixture-of-experts arm ipad

All captures:

Full page

Captured 1 day ago. Click for the whole screenshot.

Looks similar

Screenshot of github.com: Show HN: Write GitHub Actions in any language: Loopt

Show HN: Write GitHub Actions in any language: Loopt (github.com)

7 points by Bnjoroge 4 days ago | 0 comments

GitHub repository for Loopt, a Rust tool that lets you write GitHub Actions workflows in languages like Rust, Python, TypeScript, and Go and compiles them to YAML.

github-actions ci-cd yaml rust

Screenshot of github.com: Show HN: Ramasite, a Self-Contained Laravel/Filament/SQLite CMS

Show HN: Ramasite, a Self-Contained Laravel/Filament/SQLite CMS (github.com)

3 points by DaleFranks 1 day ago | 0 comments

GitHub repository for RamaSite, a self-contained MIT-licensed CMS built with Laravel, Filament and SQLite, largely written with Claude Code.

cms laravel filament sqlite

Screenshot of github.com: Show HN: PharmA2A, an open protocol for pharmacy-to-supplier orders on A2A

Show HN: PharmA2A, an open protocol for pharmacy-to-supplier orders on A2A (github.com)

4 points by eliefo 19 hours ago | 0 comments

GitHub repository for PharmA2A, an open-source Python protocol built on A2A for direct orders between French pharmacies and their suppliers.

a2a-protocol pharmacy ai-agents python

Screenshot of github.com: We've built an open-source feature flag server for 11 years

We've built an open-source feature flag server for 11 years (github.com)

137 points by ivarconr 7 days ago | 97 comments

GitHub repository for Unleash, an open-source (AGPL-3.0) feature management and feature flag platform with SDKs and a web dashboard.

feature-flags unleash feature-toggles continuous-delivery

Screenshot of github.com: 6-11X speedup running Qwen3.6 on my RTX 4090, full repo.

6-11X speedup running Qwen3.6 on my RTX 4090, full repo.(github.com)

10 points by AI4Science 6 hours ago | 0 comments

GitHub repository for Lightreason, a Python/CUDA project claiming LLM token-generation speedups for Qwen models on consumer GPUs, with benchmarks and a demo chat.

llm-inference speculative-decoding qwen rtx-4090

Screenshot of github.com: Cancer vaccine for Yash, he's diagnosed with stage IV pancreatic cancer

Cancer vaccine for Yash, he's diagnosed with stage IV pancreatic cancer (github.com)

23 points by AsDivyansh 2 days ago | 4 comments

GitHub repository from the RNA Foundry sharing tumour sequencing data, code, and designs for a personalized mRNA cancer vaccine for a stage IV pancreatic cancer patient.

mrna-vaccine cancer-vaccine pancreatic-cancer bioinformatics