frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Qwen3.8-27B make medium the default effort level instead of xhigh

https://github.com/alainnothere/llama.cpp/blob/disk-cache-eviction/models/templates/Qwen3.8-27B-medium-default.jinja
3•xlayn•27m ago

Comments

xlayn•27m ago
Because it's so nice to see page after page after page of... but what if... let's consider...

If you use a gguf, you can extract the template from the Qwen3.827B model using the following script https://github.com/alainnothere/llama.cpp/blob/disk-cache-ev..., use it like

ggufGetTemplate.py MODEL.gguf [OUT.jinja]

modify the template to your liking, and then use https://github.com/alainnothere/llama.cpp/blob/disk-cache-ev... to push it into the gguf.

ggufSetTemplate.py MODEL.gguf TEMPLATE.jinja [OUT.gguf]

This one https://github.com/alainnothere/llama.cpp/blob/disk-cache-ev... has the default effort set to medium, so you can just push it into your gguf and use it.

Interesting things, the "effort" is an instruction telling the model how to behave, medium doesn't have anything (at least on the version of q3827 that I have) so you can set whatever you like.

These are the defaults

low: "Reasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration."

xhigh: "Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer."

medium: nothing is set here... in the template above I set the instruction to: "Reasoning effort is set to medium. Think through the main considerations and verify the key steps, but do not be exhaustive: once the essentials check out, commit to the answer." and I like the resulting behavior.

If you wondered what happens if you tell the model to answer like a commedian, stop wondering and start experiment.

Here are the results of togling the effort per message (you have to use my version of llama.cpp in the link to try that) you can see it here: https://github.com/alainnothere/llama.cpp/blob/disk-cache-ev...

Happy hacking.

OpenAI halts testing, slows development after rogue model hacked Hugging Face

https://www.abc.net.au/news/2026-08-19/openai-slows-development-pauses-testing-after-hugging-face...
1•scorpioxy•1m ago•0 comments

She told no one about her agony except ChatGPT. What her death reveals about AI

https://www.npr.org/2026/08/18/nx-s1-5929575/ai-suicide-risks-mental-health
1•newsomix9xl•2m ago•1 comments

SpaceX tugs Starship into port after 24 days at sea

https://arstechnica.com/space/2026/08/its-christmastime-at-spacex-as-company-salvages-starship-fr...
2•Metacelsus•2m ago•0 comments

In a Heat Wave, Schizophrenia Is Much Deadlier Than Any Other Medical Condition

https://www.wired.com/story/heat-wave-schizophrenia/
1•littlexsparkee•2m ago•0 comments

Xerox invented the modern PC, then priced the Star at $16,595 (1981)

https://www.youtube.com/watch?v=a77jCSD6r_M
2•adhikaribipin•4m ago•0 comments

RF Cafe

https://www.rfcafe.com/
1•gregsadetsky•6m ago•0 comments

Show HN: Automatically detect and patch walking-dead states in Sierra games

https://github.com/katiahayati/lucasartsifier/
2•wkfauna•13m ago•0 comments

OpenLogi

https://openlogi.org/en
1•amatheus•13m ago•0 comments

Show HN: SubTrends – 18 years of Reddit data across 100K+ subreddits

https://subtrends.io/
2•dataneedscoffee•13m ago•0 comments

What's the best pitch for "Why not Claude"?

1•jlim99•14m ago•1 comments

Spinifex: Open-Source Amazon Web Services Alternative – OpenAlternative

https://openalternative.co/spinifex
1•tom_mulgadc•15m ago•0 comments

Language-like structure arises in learned signaling: evidence from birdsong

https://www.science.org/doi/10.1126/sciadv.aea3015
2•wslh•17m ago•0 comments

Tonnes of nuclear material found at Syria site, UN atomic watchdog says

https://www.rte.ie/news/middle-east/2026/0818/1588407-syria-nuclear/
3•wslh•24m ago•1 comments

Qwen3.8-27B make medium the default effort level instead of xhigh

https://github.com/alainnothere/llama.cpp/blob/disk-cache-eviction/models/templates/Qwen3.8-27B-m...
3•xlayn•27m ago•1 comments

Amazon scans and destroys books to comply with copyright law

https://bsky.app/profile/steveklabnik.com/post/3mtejjs6s2c2y
4•platzhirsch•31m ago•3 comments

Proof-Carrying Pipelines: attest local gate runs so CI can skip re-execution

https://github.com/JackCid89/proof-carrying-pipelines
1•jackcid89•32m ago•0 comments

The Vietnam Binh Chau (Chau Tan) Late Tang Wreck

https://www.koh-antique.com/client/tangwreck/tangwreck.html
2•teleforce•35m ago•1 comments

How to build an eval set you can maintain

https://langfuse.com/academy/evaluate/choosing-what-to-evaluate
2•lotteverh•40m ago•0 comments

Do All Your Agents Need Models Like Claude 5 or GPT-5.6?

https://aimoway-lab.github.io/blog/articles/do-all-your-agents-really-need-models-like-claude-5-o...
2•AIMOWAY•44m ago•0 comments

The Integer

https://gist.github.com/17twenty/1f230ab8256f42d4cd3ba931b9bd440b
8•hactually•48m ago•0 comments

That Disgraceful, Disreputable, (Wonderful) Form of Punctuation: The Parenthesis

https://lithub.com/on-that-disgraceful-disreputable-wonderful-form-of-punctuation-the-parenthesis/
5•pseudolus•49m ago•0 comments

ServiceNow are pushing customers off SaaS – are other industries doing similar?

2•rorychatt•50m ago•1 comments

Netstate – US Business Data from Primary Sources

https://netstate.co
2•qitx•53m ago•0 comments

OpenAI announces slowing pace of development after hack by rogue agent

https://www.theguardian.com/technology/2026/aug/18/open-ai-pause-hack
2•pseudolus•55m ago•1 comments

Study Finds Outdated Ideas About Antibiotics Are Common

https://healthcare.utah.edu/newsroom/news/2026/08/always-finish-your-antibiotics-longer-better-ou...
2•gumby•55m ago•0 comments

China's LandSpace beats SpaceX with steel-booster recovery after orbital launch

https://www.scmp.com/news/china/science/article/3364492/chinas-landspace-beats-spacex-steel-boost...
3•decimalenough•57m ago•1 comments

Personetta – One YAML Persona for Cursor, Copilot, Claude, and Cline

https://github.com/EdwardAF-IT/Personetta
2•EdwardAF-IT•1h ago•0 comments

Follow-Up Thoughts on Watermarking Schemes for AI-Generated Text

https://daringfireball.net/2026/08/follow-up_thoughts_on_watermarking
2•pinkmuffinere•1h ago•0 comments

Tiny satellite will use the dark side of the Moon as a shield

https://www.cam.ac.uk/research/news/tiny-satellite-will-use-the-dark-side-of-the-moon-to-eavesdro...
11•NordStreamYacht•1h ago•1 comments

The Economics and Engineering of On-Premises LLMs

https://cacm.acm.org/blogcacm/the-economics-and-engineering-of-on-premises-llms/
2•pseudolus•1h ago•0 comments