Posts

Mixture-of-Experts architecture for LLMs

Mixture-of-Experts  LLM Architecture   In a traditional "dense" AI model, every single part of the neural network activates for every single word or image it processes. A Mixture-of-Experts (MoE) architecture changes this by dividing the model into a collection of specialized sub-networks called "experts."   1. How Sparse Activation Works The gatekeeper system (called a router ) analyzes the incoming data. The router determines which specific experts are best suited to handle that data and directs the traffic only to them. If you ask a math question, the router sends the token to the "math expert." If you type a sentence in French, it activates the "French language expert." The remaining experts stay completely inactive ("sparse") for that token. 2. Dense vs. MoE (Sparse) Imagine a 100-billion-parameter model: Dense Model: All 100 billion parameters fire for every single word. This requires an immense amount of computing power (FLOPs...
Post a Comment
Read more

Physical Server for USGenWeb Archives

Physical Server  for USGenWeb Archives To handle an enterprise-grade workload (2,500 concurrent users and 2 million files) with 4 TB of storage, you should look at enterprise infrastructure rather than regular desktop PCs. Regular computers lack the dual power supplies, error-correcting (ECC) memory, and Linux-certified hardware required to run 24/7 without crashing under load.   I can either lease a pre-configured server from a data center (strongly recommended for high traffic) or buy the physical hardware to rack yourself. The choices are: Option 1: Live-Provisioned Dedicated Servers (Leasing) Leasing a server in a data center guarantees high-speed network connections, power backups, and immediate replacement if hardware fails. ReliableSite: You can deploy high-performance setups like the ReliableSite AMD Epyc Dedicated Servers , which offer 16-core platforms starting with 64GB of DDR5 RAM. You can use their server customizer to easily upgrade the baseline NVMe storage t...
Post a Comment
Read more

Terraform IaC

Image
  Terraform IaC
Post a Comment
Read more

Oracle NetSuite

 Oracle NetSuite SuiteScript 2.x (Core Custom Code)  What it is: NetSuite’s proprietary JavaScript-based API framework. How it works: It allows developers to write server-side scripts and client-side logic to alter system behavior. Key feature: Version 2.x is a modern, modular, architecture-driven framework that replaced the older 1.0 version. It uses specific script types (User Event, Scheduled, Map/Reduce, Client scripts) to trigger code exactly when data is saved, loaded, or processed in bulk. SuiteFlow (Visual Workflows) What it is: A point-and-click graphical workflow tool. How it works: It allows business analysts and developers to automate business processes (like multi-stage invoice approval routing) without writing heavy code. Key feature: Developers often combine SuiteFlow with SuiteScript. When a workflow hits a complex business logic barrier that the visual editor cannot handle, a developer will inject a custom SuiteScript action into the flow. SuiteTalk / RESTl...
Post a Comment
Read more

ReAct and Reflection in AI

ReAct and Reflection in AI ReAct: The “Think and Do” Agent The ReAct (Reason + Act) framework gives AI a structure for interacting with the world. It’s powerful for tasks where an AI needs to use tools, fetch data, or navigate environments. But here’s the catch: once it produces an answer, there’s no mechanism to check if it was wrong. Reflection: The “Learn From Mistakes” Agent Instead of stopping after one answer, a reflective agent has a feedback loop. Reflection gives AI something like metacognition — the ability to think about its own thinking. The tradeoff is that it’s slower and more expensive (more processing steps).
Post a Comment
Read more

Infrastructure as Code

Infrastructure as Code    Infrastructure as Code (IaC) tools like Terraform automate the deployment and management of cloud resources. The main alternatives to Terraform fall into three main categories based on their architecture and language support: Cloud-Agnostic Tools (Code-Based) 1) Terraform: Gives Multi-Cloud Support. uses plugins called "providers". uses HashiCorp Configuration Language (HCL) (ASCII, JSON Compatible, proprietary tags) to define the desired end-state of your infrastructure. 2) Pulumi: Uses familiar programming languages like Python, TypeScript, and Go instead of domain-specific languages (DSL) like HCL. 3) OpenTofu: An open-source, community-driven fork of Terraform that functions as a direct drop-in replacement. Cloud-Native Tools (Cloud-Specific) 1) AWS CloudFormation: The native declarative framework specifically built for managing Amazon Web Services infrastructure. 2) Azure Bicep: A domain-specific language designed to deploy Microsoft Azur...
Post a Comment
Read more

AIs: LangGraph, Ollama, etc.

AIs: LangGraph, Ollama, etc.     LangGraph is the orchestration harness for LangChain agents.     Ollama is an open-source tool and framework designed to run, manage, and orchestrate various LLMs locally.  Ollama allows you to download and run dozens of different open-weight LLMs, including Meta's Llama, Alibaba's Qwen, Google's Gemma, and Microsoft's Phi. Ollama requires LLMs to be quantized into the GGUF format (the format used by llama.cpp) and configured with a Modelfile. Ollama does not fine-tune LLMs.   LLMs - usually in safeTensor or .gguf file format.  The .gguf file format is for local use.   Fine-Tuning =>  LLaMA-Factory is currently the most popular tool for local fine-tuning, because it provides a complete Web UI. You do not need to write any code to use it. It supports exporting trained models directly into the GGUF format needed for Ollama.   The NVIDIA DGX Spark  (custom Linux OS) and  R...
Post a Comment
Read more