← AI news, last 7 days

llama.cpp and ggml 1.0

Last scan Sat Nov 21, 06:54 UTC
2 sources · tracked since Nov 14 · single origin (llama.cpp (ggml-org))

Version 1.0 has now been reached by both llama.cpp and ggml. According to the AINews newsletter from Latent Space, the release signals that the projects are stable, and it also drops 35 deprecated functions from the public header of llama.cpp.1 Hugging Face and ggml-org announced the v1.0.0 release, saying they intend to keep supporting a wide range of platforms and model types while ensuring the runtime stays "robust and lean" for downstream integrations.2 Among the new features mentioned: people who run mixture-of-experts models that exceed their GPU's memory can now reserve a cache for the experts they use most often.3

Hugging Face also announced Llama desktop apps built on llama.cpp, intended to cut local model setup to a few clicks.4 The newsletter says SGLang also announced a 1.0 release, though its main story that day was a separate "No AI Slop" manifesto.5

The main concern raised in comments is maintainership: commenters say llama.cpp has many open issues and PRs and ask for more maintainers.6 Other commenters emphasize its practical value as a hands-on way to start experimenting with local models.7

Every sentence was checked word for word against quotes from the articles it cites. Each footnote shows its quote, unless the page already quotes as much of that article as we allow; then it links the article instead. A model wrote and arranged the sentences, and a second read checked each against its quotes (written 2026-11-20).

Sources quoted

  1. Congrats to both llama.ccp and ggml for reaching 1.0 (declaring stability and removing 35 deprecated functions from llama.cpp’s public header) — Latent Space, [AINews] The No AI Slop Manifesto
  2. llama.cpp v1.0.0 has been announced by Hugging Face/ggml-org, with a stated goal of continuing broad platform and model-type coverage while keeping the runtime “robust and lean” for downstream integrations — Latent Space, [AINews] The No AI Slop Manifesto
  3. Users running mixture-of-experts models larger than their GPU memory can now set aside a cache for frequently used experts. — Latent Space, [AINews] The No AI Slop Manifesto
  4. Hugging Face also announced Llama desktop apps built on top of llama.cpp , intended to reduce the setup barrier for running local models to a few clicks. — Latent Space, [AINews] The No AI Slop Manifesto
  5. with SGLang also announcing 1.0 at the recent SGLang Summit. However, today’s title story goes to the No AI Slop manifesto — Latent Space, [AINews] The No AI Slop Manifesto
  6. Quote not reprinted: this page already repeats as much of this article as we allow. — Latent Space, [AINews] The No AI Slop Manifesto
  7. Other comments emphasize the project’s practical impact as a hands-on entry point for local model experimentation. — Latent Space, [AINews] The No AI Slop Manifesto

Who reported it

Facts and takes

Facts: 3 checked facts from 2 sources

Timeline

Built 2026-11-21 06:54 UTC. Facts and quotes say who states them, not that they are true.