New pages

1 July 2026

• 21:5721:57, 1 July 2026 Qdrant (hist | edit) [98,783 bytes] Wikisailor (talk | contribs) (Created page with "==Introduction== '''Qdrant''' stands as a high-performance, production-ready vector similarity search engine and database, meticulously engineered to address the demanding requirements of modern generative AI applications. At its core, it functions as a robust storage and retrieval system for high-dimensional vectors, which are essential for powering the semantic search and context-retrieval pipelines integral to systems like '''Glossary#RAG|Retrie...")
• 20:4420:44, 1 July 2026 Hierarchical Navigable Small World (hist | edit) [56,645 bytes] Wikisailor (talk | contribs) (Created page with "==Introduction== Hierarchical Navigable Small World (HNSW) is a cutting-edge graph-based algorithm utilized for Approximate Nearest Neighbor (ANN) search in high-dimensional vector spaces. Within the broader ecosystem of the '''Glossary''' infrastructure, it functions as the critical indexing mechanism that enables real-time retrieval from massive vector databases. Unlike exhaustive search methods that compare a query vector against every single data point—a proce...")
• 19:4419:44, 1 July 2026 Perplexity (hist | edit) [25,878 bytes] Wikisailor (talk | contribs) (Created page with "==Introduction== Perplexity is a fundamental metric used to evaluate the performance of probabilistic models, most notably '''Large Language Models'''. In its simplest form, it provides an objective measure of how well a probability distribution—or a model—predicts a specific sample of text. If you imagine the model as an agent attempting to guess the next word in a sentence, perplexity acts as a measure of the agent's uncertainty. A model with low...")
• 17:0617:06, 1 July 2026 High-Bit Quantization (hist | edit) [54,461 bytes] Wikisailor (talk | contribs) (Created page with "==Introduction== '''High-Bit Quantization''', specifically targeting FP16 (16-bit floating point) and BF16 (Brain Floating Point), serves as the bedrock for modern, efficient deep learning deployment. While full-precision training (FP32) is often required for the stability of initial model convergence, these high-bit formats provide the necessary balance between computational throughput and output fidelity for '''Glossary...")
• 16:0316:03, 1 July 2026 Paged Attention (hist | edit) [18,840 bytes] Wikisailor (talk | contribs) (Created page with "==Introduction== PagedAttention represents a fundamental shift in how modern inference engines manage memory for Large Language Models by addressing the fragmentation and inefficiency inherent in traditional KV Cache management. By moving away from static memory allocation, this technique allows for the non-contiguous storage of KV Cache blocks. This approach draws inspiration from...")
• 15:0115:01, 1 July 2026 NVIDIA Tensor Cores and AMD Matrix Cores (hist | edit) [11,208 bytes] Wikisailor (talk | contribs) (Created page with "== Introduction == This reference page outlines the hardware acceleration components that drive our local machine learning infrastructure on nodes like Arbour and Grape and our experimental unified memory systems. Understanding the operational distinctions between general purpose scalar processing units and specialized matrix acceleration hardware is essential for optimizing our inference pipelines and managing hardware performance expectations. Wh...")
• 13:4213:42, 1 July 2026 Asymmetric Quantization (hist | edit) [10,823 bytes] Wikisailor (talk | contribs) (Created page with "== Introduction == This guide provides a comprehensive technical breakdown of Asymmetric '''Quantization''', an advanced numerical compression technique fundamental to deploying massive neural networks on our local infrastructure. While standard quantization broadly reduces the physical memory footprint of a '''Large Language Model''', understanding the precise mechanics of asymmetric scaling reveals how w...")
• 09:3609:36, 1 July 2026 Large Language Model Introduction (hist | edit) [41,010 bytes] Wikisailor (talk | contribs) (Created page with "== Introduction == Welcome to the Large Language Model reference ledger for the Neural Systems Engineering project. This comprehensive guide details the foundational architecture, active model families, and specific hardware orchestration strategies applied across our local infrastructure network. Large language models represent the core cognitive engine of modern artificial intelligence, driving everything from automated terminal execution loops to semantic search clust...")

27 June 2026

• 23:4923:49, 27 June 2026 Scaled Dot-Product Attention (hist | edit) [8,081 bytes] Wikisailor (talk | contribs) (Created page with "==Introduction== This guide provides a comprehensive technical breakdown of Scaled Dot-Product Attention, the core mathematical mechanism that drives the reasoning capacity of modern Large Language Models (LLMs). While our memory management pages detail how parameters occupy space, this manual explains the actual linear algebra occurring inside our GPUs during the prefill and decode execution loops. The ability of a 14B model to isolate a subtle stack trace out of a mas...")
• 19:2919:29, 27 June 2026 Understanding Token and Cache Mechanics (hist | edit) [20,056 bytes] Wikisailor (talk | contribs) (Created page with "==Introduction== This guide maps the foundational memory architecture of local Large Language Models by examining the relationship between Tokens, the Context Window, and the Key-Value (KV) Cache. In local infrastructure development, memory management is not an abstract concept; it is a physical constraint dictated entirely by GPU hardware boundaries. Understanding how text is parsed, how a model maintains its train of thought, and how VRAM is consumed dynamically durin...")
• 17:3017:30, 27 June 2026 Understanding Quantization Mechanics (hist | edit) [11,345 bytes] Wikisailor (talk | contribs) (Created page with "==Introduction== This guide provides a comprehensive reference for Quantization Mechanics, the primary data compression framework that enables high-capacity Large Language Models to run locally within our infrastructure. By reducing the numerical precision of a model's parameters, we can compress its physical memory footprint by nearly 75% with negligible loss to its logical reasoning capabilities. This page details the underlying mechanics of this compression, maps the...")
• 14:5114:51, 27 June 2026 Inference Hardware Guide (hist | edit) [16,071 bytes] Wikisailor (talk | contribs) (Created page with "==Introduction== This guide maps LLM parameter sizes to available GPU hardware. It provides a baseline for selecting models that balance reasoning capability against the specific VRAM constraints of your lab's hosts, including the P4000, 5060Ti, and 4080 Super. Refer back to the '''Glossary''' for core definitions.")
• 14:3514:35, 27 June 2026 Engineering Foundations (hist | edit) [753 bytes] Wikisailor (talk | contribs) (Created page with "Introduction This page defines the core technical and philosophical parameters of the Neural Systems Engineering project. It serves as the baseline for all subsequent development, housing the essential terminology, hardware requirements, and design standards that govern the laboratory's cognitive infrastructure. By establishing these foundations, we ensure that the integration of local LLMs and agentic services remains robust, scalable, and fully under local control...")
• 14:1814:18, 27 June 2026 Glossary (hist | edit) [57,897 bytes] Wikisailor (talk | contribs) (Created page with "==Introduction== This page serves as the foundational reference for the Neural Systems Engineering project. It provides a standardized set of definitions for the technical terminology, acronyms, and operational concepts used throughout the lab's AI documentation. By maintaining a consistent lexicon, this glossary ensures that all infrastructure development, from hardware optimization to agentic scripting, is documented with precision and clarity. Refer to this list to r...")
• 14:0814:08, 27 June 2026 Neural Systems Engineering (hist | edit) [1,151 bytes] Wikisailor (talk | contribs) (Created page with "==Introduction== This is the central hub for the Neural Systems Engineering project. This section documents the design, deployment, and operation of a sovereign, locally-hosted artificial intelligence environment. Moving beyond standard homelab practices, this infrastructure integrates high-performance LLM inference, vector-based memory retrieval, and agentic automation to provide a self-contained cognitive layer for the network. ==Engineering Foundations== This se...")

25 June 2026

• 16:2216:22, 25 June 2026 2026 Proxmox Servers (hist | edit) [21,207 bytes] Wikisailor (talk | contribs) (Created page with "==📋Introduction== After some time passed and we decided to pursue the AI technology trend we decided that the ''' original server layout''' was not well suited to the new and changed role. Pear was a general purpose Proxmox host with a fast CPU and Maximum Ram and three 16 TB Ironwolf Pro HDDs making a great system for a home lab with only one host. As time passed the quantity of virtual machines increased the host became bogged down with low dema...")

11 May 2026

• 21:1521:15, 11 May 2026 Minecraft (hist | edit) [21,482 bytes] Wikisailor (talk | contribs) (Created page with "==Introduction== We are to have some Minecraft servers on the Homelab. The general setup will be to have a Velocity server in front several Minecraft host VMs. Velocity works in a similar way to a Nginx reverse proxy whereby Pfsense forwards all Minecraft traffic to the Velocity server including the SNI and Velocity redirects traffic to the required hostname based on the DNS name. Cloudflare does not reverse proxy Minecraft traffic on the free tier and as we do not...")

7 April 2026

• 15:2615:26, 7 April 2026 Silvan Reporter (hist | edit) [1,519 bytes] Wikisailor (talk | contribs) (Created page with "==Introduction== The Intention of the Silvan Reporter is to generate reports on the current state of the Homelab. it will also be able to show the current status of all of the entire network. ==Outline== The outline of the project is to have a LXC on Orchard with a SWI Prolog orchestrator with a Go application that builds a model of the current Silvan network then uses the data from the Victoria metrics / Prometheus application to report on the current status of all o...")