Articles

Deploying multiple domain-specific small language models can be challenging. This article focuses on designing an architecture that helps maintain a fleet of SLMs, covering topics such as model registries, gateway patterns, and manifest-based delivery systems.

Large Language Models (LLMs) introduce operational complexity that traditional monitoring approaches were never designed to handle. This article shows how to build end-to-end LLM observability in a FastAPI application using OpenTelemetry.

GPT-3's impressive performance revealed a limitation: raw capability doesn't automatically create a useful assistant. Researchers focused on teaching language models to better follow human intent, leading to the development of instruction-following systems and Reinforcement Learning from Human Feedback (RLHF).

Researchers at Google Research introduced Chain-of-Thought Prompting, a technique that elicits reasoning in large language models. By guiding models to generate intermediate reasoning steps, they demonstrated that many reasoning capabilities can be unlocked through prompting alone.

HyDE uses a language model to generate hypothetical documents and an unsupervised dense retriever to map those documents into an embedding space, improving the accuracy of RAG systems.

Managing AI models, datasets, and labels can be a daunting task. The 4-stage AI asset lifecycle offers a structured approach to asset management, covering creation, deployment, maintenance, and retirement.

Selecting the right Large Language Model (LLM) for your GenAI application is critical, as different LLMs can behave very differently for the same prompt. This article will walk you through a step-by-step methodology to evaluate and select an LLM for your application.

When it comes to local language models, choosing the right quantization method can significantly impact performance. In this article, we'll compare 4-bit and 8-bit quantization, exploring quality benchmarks, speed improvements, and VRAM savings.