ZML, an emerging French artificial intelligence startup, has unveiled new software called ZML/LLMD, which promises to significantly enhance the performance of large language models (LLMs) across a diverse array of AI chips. Endorsed by Turing Award laureate Yann LeCun, this release marks a potential shift in how businesses and cloud providers manage the computational demands of AI, offering a path to greater efficiency and potentially lower operational costs.
The ZML/LLMD software is designed to optimise 'inference' – the crucial process where AI models interpret and respond to prompts – across a wide spectrum of hardware. This includes chips from industry giants like Nvidia, AMD, Google's TPUs, Apple Metal, and Intel Arc. According to ZML founder Steeve Morin, the ambition is to dismantle existing hardware 'silos' and allow various chips to operate at their peak performance for AI applications, and in some cases, even faster than before. This flexibility could be a game-changer for organisations grappling with the escalating costs associated with deploying and running AI.
Morin highlighted that as AI becomes increasingly integrated into daily operations and consumer services, optimising inference has gained paramount importance, often surpassing the significance of model training. He noted that current software and architectural barriers frequently lead to vendor lock-in, limiting choices and driving up expenses. ZML's solution aims to empower enterprises and cloud providers to combine different types of chips, potentially selecting more cost-effective or energy-efficient options to build their AI infrastructure.
While ZML/LLMD is not open source, unlike some of the company's previous projects, it is being launched as a free product. This strategy allows ZML to gather valuable usage data and understand market needs before considering a revenue generation model. Morin stated that this approach prioritises growth and widespread adoption over immediate profitability, indicating a long-term vision for the product's impact on the AI ecosystem.
The initiative could also provide a significant boost to novel AI chipmakers, many of whom are based in Europe. Morin cited several European companies in this space, suggesting that ZML's technology could unlock new possibilities for hardware innovation. Despite this, ZML maintains a positive relationship with Nvidia, acknowledging its current market position and supply capabilities, while simultaneously working towards a more diversified and competitive AI hardware landscape.