frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Show HN: Getting GLM 5.2 running on my slow computer

https://github.com/JustVugg/colibri
63•vforno•13h ago
A few days ago I found myself trying out GLM 5.2 and was really positively impressed. The capabilities and security I was getting from this LLM are similar to those I've gotten from models like Claude or GPT, and this really surprised me.

But then I thought, "I wonder how it would work on a normal computer like mine," and above all, "I wonder if it would work without going into OOM on a computer like mine." So I started working with the help of agents to test this possibility.

I started converting the model to int4, understanding MTP usage, and if possible implementing DSA for long context. How it responds in int4 and whether the quality is maintained or not. Until I got to the point, on my computer with 32GB of RAM, I was able to communicate with GLM 5.2 with times that, of course, aren't high in cold start, but even then, we're talking about 0.1 tok/s, but that wasn't important to me. The important thing was the journey to reach this goal. I just wanted it to work at all costs, even slowly.

So I created Colibrì, which was born from a very simple idea, to be honest, but tested in every way, where a 744B Mixture-of-Experts model activates only ~40B parameters per token—and only ~11 GB of those change from token to token (the routed experts). So:

The dense part (attention, shared experts, embeddings—~17B params) stays resident in RAM at int4 (~9.9 GB); The 21,504 routed experts (75 MoE layers × 256 experts + the MTP head, ~19 MB each at int4) live on disk (~370 GB) and are streamed on demand, with a per-layer LRU cache, an optional pinned hot-store, and the OS page cache as a free L2.

The engine is a single C file (c/glm.c, ~1,300 lines) plus small headers. No BLAS, no Python at runtime, no GPU.No GPU or serious hardware because I don't have that hardware so I can't test it on hardware that is more powerful than my computer.Colibrì is a one-person project, written and tested entirely on a 12-core laptop with 25 GB of RAM — the numbers above are the ceiling of what I can measure at home.

Any feedback is welcome! (and if anyone wanted to participate in the project I would be delighted)

Repo: https://github.com/JustVugg/colibri

Comments

miohtama•26m ago
This is the hacker spirit
vforno•24m ago
Thank you so much, it's true! It all started with this spirit!
nerder92•21m ago
Is this inspired by antirez work on ds4?

Amazing job!

vforno•18m ago
Antirez is the number one!thanks really thanks!
Pragmata•15m ago
Would this cause issues with SSD lifespan?
vforno•13m ago
What causes problems is the rewriting in this case are only read while writing is the cache! However, I'm working to improve more and more and make some parts lighter!
mariopt•14m ago
I wonder if you could replicate this in a Colourful GeForce RTX 50-series GPU, they ship it with 2 NVMe drive slots.
vforno•12m ago
I'd love to! Right now I only have a very consumer-grade computer that I've had fun with! We'll see!
walrus01•12m ago
My main question is whether when put into practical use, this can be measured in tokens/second, or more like 1 token per minute... I have seen locally hosted LLM that are as slow as 1 tok/second still be very useful if you give it a project to do something overnight and metaphorically walk away from it, check back with what it has done in 6 or 8 hours.

0.05 to 0.1 tok/s on the other hand, as reported in the URL for the lowest class of hardware, isn't really usable for much.

vforno•11m ago
In the readme you can see benchmark which everyone with different hardware is running Colibrì, and I have to say I've seen great times! I'm always doing more to improve!
kzrdude•11m ago
Your coding style is halfway to IOCCC. I'm just jealous though :)

GPT-5.6

https://openai.com/index/gpt-5-6/
796•logickkk1•4h ago•590 comments

Show HN: Getting GLM 5.2 running on my slow computer

https://github.com/JustVugg/colibri
72•vforno•13h ago•20 comments

EU Parliament greenlights Chat Control 1.0

https://www.patrick-breyer.de/en/eu-parliament-greenlights-chat-control-1-0-breyer-our-children-l...
800•rapnie•10h ago•394 comments

Show HN: 18 Words

https://18words.com/
715•pompomsheep•8h ago•260 comments

Buried Apple feature turns an iPhone into the perfect kids' dumb phone

https://www.wired.com/story/this-buried-apple-feature-turns-an-iphone-into-the-perfect-kids-dumb-...
177•PotatoNinja•3d ago•115 comments

Hy3

https://hy.tencent.com/research/hy3
288•andai•5h ago•66 comments

Train SIM Created by Just One Person Is Being Called the Best Ever Made

https://kotaku.com/a-train-sim-created-by-just-one-person-is-being-called-the-best-ever-made-2000...
94•oumua_don17•4d ago•34 comments

No leap second will be introduced at the end of December 2026

https://datacenter.iers.org/data/latestVersion/bulletinC.txt
191•ChrisArchitect•7h ago•154 comments

I Changed My Name

https://robida.net/entries/2026/07/01/i-changed-my-name
14•surprisetalk•3d ago•1 comments

GLM 5.2 is nearly as accurate as a human book keeper

https://toot-books.pages.dev/blog/glm-5-2-vat-benchmark
134•adamkurkiewicz•2h ago•81 comments

A road to Lisp: Why Lisp

https://scotto.me/blog/2026-07-09-why-lisp/
62•silcoon•8h ago•61 comments

Girls just wanna have fast MPMC queues with bounded waiting

https://nahla.dev/blog/waitfree_queue/
93•EvgeniyZh•3d ago•16 comments

A possible future for Damn Interesting

https://www.damninteresting.com/a-possible-future/
167•mzur•5h ago•15 comments

ChatGPT Work

https://openai.com/index/chatgpt-for-your-most-ambitious-work/
285•Tiberium•4h ago•125 comments

Wildcard (YC W25) Is Hiring a Founding Engineer

https://www.ycombinator.com/companies/wildcard/jobs/ZSLVaaU-founding-engineer
1•kaushikmahorker•4h ago

The glass backbone: Why the Army's logistics will break in the next war

https://mwi.westpoint.edu/the-glass-backbone-why-the-armys-logistics-will-break-in-the-next-war/
231•baud147258•7h ago•292 comments

Why the Next Era of AI Is About Infrastructure, Not Just Models

https://blog.mozilla.ai/the-control-layer-why-the-next-era-of-ai-is-about-infrastructure-not-just...
24•royapakzad•6h ago•9 comments

TLS certificates for internal services done right

https://tuxnet.dev/posts/tls-for-internal-services/
104•mrl5•6h ago•69 comments

How to Start a Ruby Meetup

https://guides.rubyevents.org/meetups/
40•mooreds•2h ago•10 comments

Muse Spark 1.1

https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/
276•ot•7h ago•157 comments

Postgres rewritten in Rust, now passing 100% of the Postgres regression tests

https://github.com/malisper/pgrust
186•SweetSoftPillow•14h ago•238 comments

How GitHub gave every repository a durable owner

https://github.blog/security/application-security/how-github-gave-every-repository-a-durable-owner/
6•ascertain•29m ago•0 comments

Opinionated and Easy Pi.dev Configuration

https://lazypi.org/
84•lwhsiao•5h ago•53 comments

Launch HN: Context.dev (YC S26) – API to get structured data from any website

https://www.context.dev
56•TheYahiaBakour•5h ago•41 comments

Almost Always Unsigned

https://graphitemaster.github.io/aau/
32•gavide•1d ago•45 comments

AI content is everywhere on social media, especially LinkedIn

https://www.pangram.com/blog/ai-in-your-feed
128•mukmuk•5h ago•113 comments

Show HN: I built a web tool to see and edit what an AI thinks before it answers

https://lucid.earthpilot.ai
4•ada1981•3h ago•0 comments

Show HN: Sighthound - open-source vulnerability scanner for source code

https://github.com/Corgea/Sighthound
4•asadeddin•5h ago•0 comments

Show HN: I mapped 8.5M research papers into an interactive atlas

https://tomesphere.com/atlas
49•leonickson•18h ago•15 comments

Building a real-time AI tutor for 5-year-olds

https://www.ello.com/blog/teaching-a-child-in-1000-ms
5•catalinvoss•25m ago•1 comments