Large-language-models

This post explores RAPTOR, a novel approach to retrieval-augmented language models that constructs a hierarchical tree structure of documents through recursive embedding, clustering, and summarization. This method enables retrieval of information at different levels of abstraction, significantly improving performance on complex question answering tasks involving long documents compared to traditional contiguous chunk retrieval.

Discover the definitive guide to large language models (LLMs) with the book Large Language Models: A Deep Dive. This comprehensive review explores how the book bridges cutting-edge AI theory with real-world applications, offering unparalleled insights into the latest advancements, practical use cases, and ethical considerations in the field of LLMs.

Discover how QwQ-32B, a 32-billion-parameter model, leverages reinforcement learning to achieve state-of-the-art performance in reasoning and tool utilization, rivaling models with significantly larger parameter counts.

Explore how reinforcement learning and large language models like OpenAI's o3 are transforming competitive programming, surpassing specialized systems without relying on hand-crafted strategies.

This paper challenges the conventional wisdom of mixing different Large Language Models (LLMs) in ensemble methods. It introduces Self-MoA, a novel approach that aggregates outputs from only the top-performing LLM, and demonstrates its superiority over standard Mixture-of-Agents (MoA) in various benchmarks.