TAGGED GUIDES
Explore the guides, ideas, and practical explanations in this topic.
18 published articlesA single-GPU RAG stack works once you decide which process owns the GPU. The most predictable pattern reserves the GPU for generation and keeps embeddings and retrieval…
WP-CLI handles the deterministic WordPress operations, and a local Ollama model supplies the judgement layer that shell scripts lack. This guide shows how to pipe WP-CLI's JSON…
Exact reproduction of an Ollama run requires pinning five layers at once: the model artifact, the runtime version, the sampling parameters, the prompt and context, and the…
Ollama supports two distinct capabilities that are easy to confuse: tool calling, where a model requests a named function that your app executes, and structured outputs, where…
A local AI runtime such as Ollama, LM Studio, or llama.cpp usually ships with no authentication and a plain-HTTP listener bound to localhost. WordPress should never talk…
A retrieval-augmented generation pipeline can run entirely on your own machine, including the embedding step that turns documents into vectors. This guide maps the local-only component stack,…
A local RAG pipeline uses Ollama to serve both the chat model and the embedding model, and ChromaDB to store and search document vectors, so answers are…
A hallucination evaluation for a local model needs three things: questions whose answers you have verified, a cold run with no tools or retrieval, and a grading…
Local LLM endpoints time out for reasons cloud APIs rarely do: cold model loads, KV cache allocation, prompt evaluation, and queueing behind other requests. This guide explains…
A fixed test suite replaces impressions with a comparable number: freeze the tasks, the harness, the runtime, and the scoring rule, then run the same suite repeatedly.…
Local AI means running a language model on hardware you control instead of sending prompts to a hosted service. This guide explains how local inference works, how…
Raising a local model's context window from 2K to 32K changed generation speed by essentially nothing in a published llama.cpp sweep, but added about 1.7 GB of…