RAG-Native LLMs, Homogeneous Clusters, and Cheap RLHF

Tools

Exo Now Supports Distributed LLM Inference Across Heterogeneous Devices

Exo allows you to build a 'homogeneous' cluster from heterogeneous devices (phones, Macs, old GPUs) by abstracting away the hardware differences. This is the first real tool for creating a personal compute cloud for massive models.

AutoRAG Launches a Local, No-Code Pipeline for Building Private RAG Datasets

This tool lets you take massive datasets, chunk them intelligently, and index them into a vector store for RAG without ever sending a byte of data to the cloud. It is a fully self-contained, high-throughput pipeline.

News

Alibaba Drops Qwen2.5-Turbo with "Native" RAG Architecture

Designed specifically for RAG and grounded generation, this model uses a unique 'thinking-retrieval-generation' loop that is far more honest about its sources than standard RAG setups.