LLaMA now goes faster on CPUs(justine.lol)

1286 points by lawrencechen 2 years ago | 413 comments

Justine Tunney's blog post describing 84 new matrix multiplication kernels that make llamafile prompt evaluation 30-500% faster on CPUs than llama.cpp.

QuIP#: 2-bit Quantization for LLMs(cornell-relaxml.github.io)

185 points by jasondavies 2 years ago | 41 comments

I added some optimizations to my compiler that turns Lisp into JavaScript(healeycodes.com)

102 points by healeycodes 2 years ago | 17 comments

How Is LLaMa.cpp Possible?(finbarr.ca)

648 points by birriel 3 years ago | 210 comments

Blog post by Finbarr Timbers explaining the math of why llama.cpp can run LLaMA models locally on consumer hardware, focusing on memory bandwidth and compute limits.

Page Table Memory Consumption(frn.sh)

57 points by shellpipe 25 days ago | 10 comments

Blog post by Fernando Simões examining page table memory consumption, prompted by Linus Torvalds' 2003 argument for multi-level over hashed page tables.

Turbocall – Just-in-time compiler for Deno FFI(divy.work)

80 points by undefined_void 2 years ago | 10 comments

AI Made Writing Code Easier. It Made Being an Engineer Harder(ivanturkovic.com)

343 points by saikatsg 7 months ago | 242 comments

Blog post by Ivan Turkovic arguing that AI tools make producing code easier while making the broader job of being a software engineer harder.

JavaScript Demos in 140 Characters(beta.dwitter.net)

325 points by themanmaran 9 months ago | 59 comments

Dwitter, a community site for sharing and remixing JavaScript canvas demos written in 140 characters or less.

That's not an abstraction, that's a layer of indirection(fhur.me)

546 points by fagnerbrack 1 year ago | 213 comments

A blog post arguing that many so-called abstractions in codebases are merely layers of indirection that add complexity and hurt performance without real value.

How to setup a local coding agent on macOS(ikyle.me)

464 points by kkm 4 months ago | 114 comments

A blog tutorial on running Gemma 4 and Qwen3.6 locally on macOS with llama.cpp, MTP speculative decoding, and multimodal support as a coding agent.

I built a programming language that LLMs can't use(blog.picheta.me)

244 points by dom96 8 days ago | 179 comments

Blog post by Dominik Picheta introducing KillSwitch, a programming language he designed so that large language models cannot use it.

I don't like NumPy(dynomight.net)

399 points by MinimalAction 1 year ago | 164 comments

A Dynomight blog essay critiquing NumPy's design, arguing its array broadcasting and indexing make complex computations confusing.