Microsoft Deploys Custom Maia 200 Chip To Reshape Cloud AI Economics

The Maia 200 deployment demonstrates that custom silicon has matured from experimental capability to production ...
IEEE

LLM on FPGA: Squeezing Language Models by Quantization and Multi-Query Attention and its Efficient Hardware Architecture

Abstract: We present an on-chip implementation of a compressed Transformer-based language model on a Xilinx Artix-7 FPGA. Our contributions include: (1) combining ultra-low-precision quantization (4 ...
IEEE

Enhancing Generative Models with Hierarchical Architecture for Automated Product Advertising Content Generation

Abstract: Automated product advertising content generation aims to create compelling and persuasive ads to attract users. While various approaches exist to generate ad content, we adopt a ...

AI systems learn better when trained with inner speech like humans

Talking to yourself feels deeply human. Inner speech helps you plan, reflect, and solve problems without saying a word.
The Official Microsoft Blog

Maia 200: The AI accelerator built for inference

Today, we’re proud to introduce Maia 200, a breakthrough inference accelerator engineered to dramatically improve the economics of AI token generation. Maia 200 is an AI inference powerhouse: an ...