frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac

https://github.com/drumih/turbo-fieldfare
132•gitpusher42•1h ago•27 comments

KOReader

https://koreader.rocks/
463•Cider9986•5h ago•152 comments

Handbook.md shows that long policy documents do not reliably govern agents

https://arxiv.org/abs/2607.25398
169•spIrr•3h ago•99 comments

Hugging Face: Anatomy of a frontier-lab agent intrusion

https://huggingface-anatomy-of-frontier-lab-model-intrusion.static.hf.space/index.html
39•dn2k•1h ago•4 comments

Launch HN: Tokenless (YC S26) – Automatic model switching to save money

https://usetokenless.com/
9•rohaga•26m ago•4 comments

Shipping Godot VR and Porting to PSVR2: A Partial Post Mortem

https://www.claire-blackshaw.com/blog/2026/07/shipping-godot-vr-and-porting-to-psvr2-a-partial-po...
56•ibobev•3h ago•0 comments

Hamburg's Stadtpark: A Park Built to Be Used

https://alsterrunde.com/hamburgs-stadtpark-a-park-built-to-be-used/
15•mertbio•2d ago•1 comments

Hunter-gatherers introduced fish to a mountain lake 7000 years ago

https://www.newscientist.com/article/2580119-hunter-gatherers-introduced-fish-to-a-mountain-lake-...
74•stevenwoo•2d ago•48 comments

Cesium DevCon 2026 talks are up, including a keynote from SQLite's creator

https://cesium.com/events/cesium-developer-conference/2026/
14•jasteinerman•1h ago•1 comments

Document-borne AI worms can self-propagate through Copilot for Word

https://enklypesalt.com/posts/context-collapse-part3-ai-worming-through-word/
202•Canopy9560•4h ago•172 comments

Darktable

https://www.darktable.org/
125•siatko•3h ago•65 comments

Self-hosting Kimi K3: 20% more hardware cost, 20% better task resolution

https://aistack.imec-int.com/blog/gpu-self-hosting
17•flifenstein•1h ago•2 comments

More Tailscale tricks for your jailbroken Kindle

https://tailscale.com/blog/jailbroken-kindle-proxy-tun-modes
342•Error6571•11h ago•100 comments

GPT-5.6 vs. Claude Fable 5 for Physical AI, which performs best?

https://juliahub.com/blog/frontier-models-physical-ai-evaluation
22•mbauman•1h ago•4 comments

Amiga Graphics Archive

https://amiga.lychesis.net/index.html
110•Bluestein•6h ago•17 comments

Disrupting supply chain attacks on NPM and GitHub Actions

https://github.blog/security/supply-chain-security/disrupting-supply-chain-attacks-on-npm-and-git...
57•nyku•4h ago•17 comments

User Interfaces of the Demo Scene

https://www.datagubbe.se/scenegui/
342•zdw•11h ago•57 comments

Superlogical – Mitchell Hashimoto

https://mitchellh.com/writing/superlogical
7•tambourine_man•36m ago•0 comments

Show HN: Qwen Scribe – local transcription and dictation for Apple Silicon

https://github.com/VladUZH/qwen-scribe
13•sidclaw•1h ago•3 comments

AI in Linux

https://drewdevault.com/blog/AI-in-Linux/
20•surprisetalk•2h ago•7 comments

SQLite in Production: Optimizing WAL Mode, Concurrency, and VFS Layers

https://micrologics.org/blog/sqlite-in-production-optimizing-wal-mode-concurrency-and-vfs-layers-...
192•ankitg12•9h ago•58 comments

SpecForge – A Platform for Authoring Formal Specifications

https://docs.imiron.io/v/0.5.10/en/tour.html
61•agnishom•5h ago•7 comments

A.I. Companies Are Recruiting Electricians and Carpenters by the Thousands

https://www.nytimes.com/2026/07/29/business/economy/data-center-electricians-training.html
54•thm•1h ago•29 comments

Show HN: Write, simulate and synthesize VHDL/Verilog in the browser

https://risingedge.pro
14•wozniakpawel•6d ago•4 comments

Ask HN: My domain registrar (Hover) rug-pulled me for $3000

12•shrinks99•35m ago•8 comments

A Texture Lookup Approach to Bézier Curve Evaluation on the GPU (JCGT)

https://jcgt.org/published/0015/02/01/
28•ibobev•3h ago•4 comments

Lisp moving Forth moving Lisp

https://letoverlambda.com/textmode.cl/guest/chap8.html
91•fallat•2d ago•22 comments

Google shuts down Nobel Prize winning AlphaFold

https://www.engadget.com/2225849/google-shuts-down-alphafold/
27•NordStreamYacht•1h ago•1 comments

Graph Engineering Needs a Compiler

https://fluxtion-playground.dev/blog/2026-07-29-graph-engineering-needs-a-compiler
8•v12technology•1h ago•0 comments

San Francisco: Don't Fall for Industry Defense of Surveillance Pricing

https://www.eff.org/deeplinks/2026/07/san-francisco-dont-fall-industry-defense-surveillance-pricing
88•hn_acker•3h ago•34 comments
Open in hackernews

Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac

https://github.com/drumih/turbo-fieldfare
132•gitpusher42•1h ago
Hi HN,

I built a specialized inference engine for running 4-bit Gemma 4 26B-A4B-IT on any M-series Mac using about 2 GB of RAM. It is called TurboFieldfare and is written in Swift and Metal.

I have always adored on-device AI. It feels like magic that you can run a powerful NN on your Mac or iPhone. So I wanted to push the limits a bit and run a model whose weights don’t fit in memory.

The model’s 4-bit quantized weights occupy roughly 14 GB, which makes running it with conventional inference tools almost impossible on an 8 GB or even 16 GB Mac once the OS, applications, and KV cache are included.

The trick is to keep the shared part of the model and the KV cache in RAM, then stream only the routed experts needed for each token from SSD. An SSD is way slower than RAM, so the runtime uses a small expert cache and bounded parallel `pread`. While those reads are in flight, the GPU runs the shared part of the layer.

I ran more than 100 experiments. Most didn’t work. A few got me here. The experiments are described in the GitHub repo.

It currently generates 5–6 tok/s on an 8 GB M2 MacBook Air and 31–35 tok/s on an M5 MacBook Pro.

I also added an experimental OpenAI-compatible local server. It supports streaming and tool calls, and reuses one prompt prefix from the KV cache.

Try it! The Mac app is easy to install. On the first run, it will download 15 GB of weights from Hugging Face. The model is surprisingly capable.

I would love any kind of feedback!

Comments

yakupov_bulat•53m ago
Wow, amazing!

What if there is enough RAM to fully load the model? I assume in that case I shouldn’t use your engine.

0gs•50m ago
you could use mine ... github.com/0gsd/enough (it has other stuff too)
gitpusher42•43m ago
It depends on the use case.

I measured this exact model with a 4k context on the mlx engine. It runs at 75 tok/s on my M5 Mac Pro and using 14 GB of RAM. For my engine the same model uses 2 GB of RAM and produces 31–35 tok/s.

The project is still experimental so performance may vary as it continues to improve. If you want to save around 12 GB of RAM for other tasks and you are ok with 35 tok/s (afaik it is roughly comparable to ChatGPT’s speed for basic responses) my engine may be a good fit.

If you need maximum speed and flexibility just use MLX

addaon•52m ago
> It currently generates 5–6 tok/s on an 8 GB M2 MacBook Air and 31–35 tok/s on an M5 MacBook Pro.

Where does this big a performance spread come from? I wouldn't naïvely expect SSD performance difference to be that big, and I would expect SSD performance to dominate...

afzalive•48m ago
The M5 MBP has 24GB of RAM, more context in RAM perhaps?
gitpusher42•21m ago
The process stays at around 2 GB with 16 slots and a 4K context on both the M5 and M2. But yeah, Apple might be doing some magic under the hood
wongarsu•46m ago
My suspicion is that this is simply due to the M5 having more memory, and the OS already having most of the file cached. The M2 has more memory pressure and would cache fewer of the SSD reads

If that's true, inference speed would be even lower if you have only 2GB total, including OS caches

GeekyBear•37m ago
The M5 SSD's performance uplift was fairly substantial, even when compared to the prior generation.

> In the Blackmagic Disk Speed Test, the SSD in the M5 MacBook Pro achieved read speeds of up to 6,323 MB/s, compared to just 2,031 MB/s on the M4 MacBook Pro. It's not like the M4 is "slow" in a vacuum, but the M5 SSD is over three times faster, which is a great generation uplift.

https://www.tomshardware.com/laptops/macbooks/m5-macbook-pro...

mxmlnkn•49m ago
This sounds really cool. My intuition was that the selected experts might change heavily for each token, resulting in slow SSD loads for each token. This seems to be wrong. Did you create some statistics on how often the experts need to be changed? What is the longest token run without any expert change? What does such a token run look like? In which cases do experts change frequently?
gitpusher42•4m ago
The full route changes almost every token. The cache works through partial reuse, about 40% of experts repeat on the next token and 57% within two tokens, cutting I/O from 166 to 88 ms/token on M2 Mac.

The longest exact repeat we found was only two tokens. Coding tasks may have higher reuse if code related experts are selected repeatedly

znpy•48m ago
I wonder if i can run this on my MacBook Neo!
gitpusher42•39m ago
I haven't tried it but it should work! You can try it and share your results, it would be really appreciated

I tried it on my wife's M1 MacBook Air 512GB and it gets 4–5 tok/s

Also, it must be easy to adjust for iPhones and iPads in theory

WithinReason•45m ago
Nice job implementing expert caching!
gitpusher42•36m ago
Thank you! Under good conditions it achieves approx a 67% cache hit rate with 16 expert slots
h2aichat•45m ago
Hope you can do it for Windows users also (and small graphics cards). Thanks
gitpusher42•34m ago
Uh, I’m afraid it is Apple only. It is written using Apple’s GPU language, Metal, and heavily relies on the Apples’s shared memory architecture

Windows PCs would require a completely different approach

touwer•37m ago
Cool! Is there any info on this doing harm to the SSD? (Or other parts?)
wtallis•36m ago
Reads don't wear out flash memory to any meaningful extent.
gitpusher42•13m ago
AFAIK it should not because it is only reading
sscarduzio•34m ago
How does this compare to DwarfStar4?
hsienchuc•26m ago
I've run local video generation models on an 8GB graphics card and know firsthand that nothing runs smoothly when memory is insufficient. So seeing 14GB of weights crammed into 2GB of RAM is impressive.

If running continuously for over an hour (like an overnight batch task), will a fanless MacBook Air overheat and throttle? Can the SSD handle the continuous weight reads and sustained output speeds?

Great work, congratulations on the release!

m00x•8m ago
This is where MoEs shine though. You don't need all experts in memory at once. Diffusion inference doesn't have sparse inference.
tredre3•25m ago
I'm curious how your project compares to plain mmap!

Because llama.cpp will already run 26B in 2GB of RAM if you really want to (mmap enabled, repacking disabled).

It seems like the main difference is that your project synchronizes the SSD reads with inference activity, which you've presumably tuned to cause the least latency possible? Whereas the OS wouldn't care about any of that.

Catloafdev•12m ago
Ya I'd be interested to see a comparison of using llamacpp with ssd offloading to compare real speeds.
greggh•22m ago
It does exactly what it says it does. On my Mac mini M4 with 16GB of ram it is running at just over 5 tok/s. That jump from M4 to M5 is crazy.
ambicapter•4m ago
Extremely impressive.