Coding Agent Adaptation Lets a 9B LLM Outperform 10x Larger Models on Aider Polyglot Benchmark

and as an output, a coding agent optimized to smaller LLMs: https://github.com/itayinbarr/little-coder

interesting that using AI models from China is not discussed.

e.g. Apple buys moonshot or z.ai

like meteorites?

Craft?

announce discussion https://news.ycombinator.com/item?id=8548429

it is called llama-barn https://github.com/ggml-org/LlamaBarn

LlamaBarn is the MacOS app, not the HTTP API server, which is "llama-server".

On non-Apple PCs, "llama-server" is what you use, and you can connect to it either with a browser or with an application compatible with the OpenAI API.

Perhaps using "llama-server" as the name of the project would have been less confusing for newbies than "llama.cpp".

I confess that when I first heard about "llama.cpp" I also thought that it is just a library and that I have to write my own program in order to implement a complete LLM inference backend.

this looks nice but is macos only.

check on same port, there is an OpenAI API https://github.com/ggml-org/llama.cpp/tree/master/tools/serv...

Good stuff, thanx!

like someone said above: brew install llama.cpp

llama-server -hf ggml-org/gemma-4-E4B-it-GGUF --port 8000 (with MCP support and web chat interface)

and you have OpenAI API on the same 8000 port. (https://github.com/ggml-org/llama.cpp/tree/master/tools/serv... lists the endpoints)

And why do I use ggml-org/gemma-4-E4B-it-GGUF instead of one of the 162 other models that can be found under the ggml-org namespace? And how do I even know that this is the namespace to look at?

That's what I meant by model management. I'm too tired to scroll through a bazillion models that all have very cryptic names and abbreviations just to find the one that works well on my system with my software stack.

I want a simple interface that a tool like me can scroll through easily, click on, and then have a model that works well enough. If I put in that much brain power to get my LLM working, I might as well do the work myself instead of using an LLM in the first place.

1. Go to HF

2. Choose the model they recommend

3. Run the one-liner the site gives you

Bonus: faster access to latest models and better memory usage

The first model I see on the HF homepage is this one: MiniMaxAI/MiniMax-M2.7

Do you think that this 229B parameter model will work on my consumer PC?

Stop pretending like HF is in any way beginner friendly.

https://www.quantamagazine.org/about/ says "launched by the Simons Foundation in 2012"

and https://www.simonsfoundation.org/about/ has "Since its founding in 1994 by Jim and Marilyn Simons"

https://en.wikipedia.org/wiki/Jim_Simons explains how Jim Simons got rich.

The book 'The Man Who Solved the Market' - https://www.gregoryzuckerman.com/the-books/the-man-who-solve... is a nice read.

HN discussion on a review of the book - https://news.ycombinator.com/item?id=29392041

it is an Ethereum fork, named after Jan Zurich (a cousin of the famous Chief Niklaus Emil Wirth). Jan Zurich discovered a little moon on Uranus, and named it Blaise

see timeline on https://ethereum.org/ethereum-forks/

It’s also worth noting that Jan (who strictly uses the pronouns var / val) belongs to one of the most historically marginalized groups in modern tech: One-Pass Compiler Enthusiasts. They were repeatedly ostracized by the bloated LLVM cabal for stating that any build process taking longer than 50 milliseconds is a toxic social construct. The ETH fork was actually meant to fund a decentralized safe space where nobody is ever forced to use a borrow checker.

related discussion https://news.ycombinator.com/item?id=40338443

Funny you should bring up MacRelix — the very first front end for Advanced Mac Substitute was built in it.