Topic
Running large language models on consumer GPUs and Apple Silicon with llama.cpp, Ollama and LM Studio.
An intern's slice of a legacy Java modernization project found the bottleneck is not generation. It is having a test that can tell you when the model is actually finished.
Reality
A hardening write-up on LM Studio separates local execution from network isolation, then enforces the second with OS rules rather than trusting the app's own checkboxes.
Reality
A developer wired runtime schema introspection, vector field matching and a local LLM repair loop into Odoo 17, then published the failure list. The failures define the boundary.
Reality
A Techdirt writer's itemized account of where AI sits in his production process is a better template for content and software teams than any yes-or-no disclosure box.
Reality
A single 8GB laptop produced four different VRAM figures, none of them a bug. The one that decides whether a model loads is the one no tool puts in front of you.
Reality
A dev.to guide argues local LLM capacity planning collapses into one napkin equation. Run it first and the hardware shortlist writes itself, tier names and all.
Reality
A dev.to writeup drops LangChain, LlamaIndex, hosted vector stores and the cloud model, leaving six files, seven npm packages, and one cloud dependency still in place.
Reality
A developer moved both routing decisions and code generation onto local models after round-the-clock agents made usage billing permanent. The CI arithmetic is published. The hardware price is not.
Reality
Two releases, two licences. Only the 27B is Apache 2.0, and because just 16 of its 64 layers keep a KV cache, long context costs a quarter of the usual memory.
Reality