frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Unsloth Dynamic 3.0 GGUFs

https://unsloth.ai/docs/basics/dynamic-3.0-ggufs
55•jonesy827•1h ago

Comments

xlayn•36m ago
Hey Unsloth, your gguf are the first ones I look for when I want to download a gguf model. Today I was trying in fact to see, what's the smallest Qwen3.8-27B that I could run and get good results, say restricting it to 16GB of ram.. so I went, pick up the Qwen3.8-27B-UD-IQ2_XXS.gguf and them BAM, error on MTP... now I understand why after reading your announcement. Beyond the space saving, why removing the MTP? improves speed exactly for the group that could benefit from it.
mike-the-brain•35m ago
you can still have it, no?

> We also removed the MTP module from smaller quants under UD-Q2_K_XL (8.37GB and lower) to converse around 500MB of disk space - you can use the Q4_0 MTP separate module if needed

xlayn•4m ago
my bad, you are totally right, thanks!
mike-the-brain•36m ago
Might be off-topic but: is it possible to perform such a quantization on Apple devices? Something like Mac Studio Ultra M1 (even if it would take weeks/months)?
throwa356262•33m ago

   "We also made some smaller UD-1bit quants with UD-IQ1_S being 6.2GB (without MTP) which retain around 72% top-1% accuracy yet being 89% smaller"

This is crazy! But has anyone tried these lower quants on real projects?
kennywinker•23m ago
Not 1-bit, but I’m getting pretty good results with some light coding using unsloth’s previous 2-bit quant of qwen3.8-27b. With these new quants i may be able to bump up to 3bit, tho it’s already running so slow (15tok/s average for the first 32k of context) that the speed hit might make it not worth the extra smarts
jadbox•27m ago
The new IQ4XS has been working pretty well so far on 4090 16gb.
kamranjon•12m ago
What size context are you able to squeeze in with less than 2gb of headroom? I have had some luck using a quantized kv cache but i fear that also decreases overall quality.
tetsuo420•13m ago
It seems the NVFP4 quants have a preview version of this Unsloth Dynamic 3.0. Is this close to the finished version, or would it be better to switch to one of the newer quants?
spwa4•10m ago
No MLX versions for 3.8 though.

OpenRouter is joining Stripe

https://openrouter.ai/blog/announcements/openrouter-is-joining-stripe/
302•rvz•2h ago•189 comments

Go 1.27

https://go.dev/blog/go1.27
192•database64128•1h ago•26 comments

Unsloth Dynamic 3.0 GGUFs

https://unsloth.ai/docs/basics/dynamic-3.0-ggufs
55•jonesy827•1h ago•10 comments

A joke domain purchase turned in geopolitical warfare

https://sprocketfox.io/xssfox/2026/08/19/sondehub-and-war/
573•kareiva•8h ago•75 comments

Pixel 11 Pro Fold feels like the end of an era

https://www.theverge.com/tech/981956/google-pixel-11-pro-fold-review
10•animalcule•27m ago•11 comments

Police officer used Flock cameras to track estranged wife 717 times

https://www.wsbtv.com/news/trending/affidavit-police-officer-used-flock-cameras-track-estranged-w...
103•speckx•1h ago•11 comments

Geolocating a random island using geometry and CUDA programming

https://yassa9.github.io/osint/gralhix-004/
339•yassa9•7h ago•57 comments

Ornith-1.5: From Self-Scaffolding to Self-Improvement

https://ornith.ai/ornith_1_5.html
127•CommonGuy•5h ago•34 comments

From Quantum Relative Entropy to the Semiclassical Einstein Equations

https://arxiv.org/abs/2510.24491
8•Luc•39m ago•0 comments

How Kubernetes Probes Work

https://ngrok.com/blog/probes
76•cyndunlop•3h ago•11 comments

Extensible Software in the age of LLMs

https://jeremymorrell.dev/blog/extensible-software-in-the-age-of-llms/
66•coloneltcb•3h ago•27 comments

Remote workers report the highest well-being in study of 7,700 employees

https://www.colorado.edu/today/2026/08/12/remote-workers-report-highest-well-being-study-7700-emp...
335•downbad_•4h ago•161 comments

Unlocking a locked/deactivated e-waste Cricut Maker

https://sprocketfox.io/xssfox/2026/07/01/cricut-unlock/
13•1e1a•1h ago•2 comments

Launch HN: OneCLI (YC S26) – OSS sandboxed agent harness for teams

https://github.com/onecli/onecli
34•guyb3•3h ago•12 comments

You Probably Don't Get Why Stripe Bought OpenRouter

https://www.amppublic.com/research/openrouter
6•amrrs•31m ago•0 comments

Rules of Good Social Skills

https://liamrosen.com/2025/07/24/33-rules-of-good-social-skills/
16•bilsbie•1h ago•2 comments

The little-known winstart.bat batch file

https://devblogs.microsoft.com/oldnewthing/20260811-00/?p=112605
7•ingve•3d ago•0 comments

Google replaced Git tags for certain source code with obtaining via Google Drive

https://grapheneos.social/@GrapheneOS/117057099753905023
38•Animux•2h ago•7 comments

Introducing MicroLighter

https://daverupert.com/2026/08/microlighter/
24•tobr•1h ago•5 comments

Mathematics in the Age of AI

https://arxiv.org/abs/2608.16753
60•jonbaer•4h ago•42 comments

Ramp Launches a Model Router

https://router.com
13•zackfield•44m ago•2 comments

Air Theremin – a browser theremin you play by waving at your webcam

https://theremin.bizibah.com/
206•gurov•9h ago•74 comments

PostgreSQL for Everything

https://www.raphaelbauer.com:443/posts/postgresql-everything/
236•karlmush•6h ago•161 comments

Moderna reports first positive Phase 3 for mRNA neoantigen therapy in melanoma

https://twitter.com/NoubarAfeyan/status/2090050162441752787
429•heydenberk•6h ago•194 comments

Chain-of-Thought Reasoning in the Wild Is Not Always Faithful

https://arxiv.org/abs/2503.08679
46•florianherrengt•3h ago•29 comments

Microgpt in pure C hits 10M tps on Apple m5

https://github.com/vixhal-baraiya/microgpt-c
98•dhorthy•1d ago•32 comments

Devices with GrapheneOS support should be available in 2027

https://grapheneos.social/@GrapheneOS/117078064184215730
533•exceptione•8h ago•341 comments

A revisit of remote Spectre attacks on Cloudflare Workers

https://blog.cloudflare.com/revisiting-spectre-attacks-on-workers/
14•albertpedersen•2h ago•0 comments

Turning molecules into reliable electronic devices

https://news.mit.edu/2026/turning-molecules-into-reliable-electronic-devices-0803
19•gmays•5d ago•0 comments

A decades-old bug in Knuth's long division (TAOCP Vol II, Algorithm 4.3.1D)

https://kolja.rs/algorithm-d/
164•nk_kolja•6d ago•38 comments