**ZML: French Startup Disrupts AI Inference with Free Multi-Chip Accelerator Software**
The era of Nvidia's unparalleled dominance in the AI chip market, while robust, is witnessing the rise of potent challengers. Among them is ZML, a burgeoning French AI startup lauded by Turing Award winner Yann LeCun, which has just unveiled its groundbreaking inference-performance software: ZML/LLMD. This newly launched LLM inference server is engineered to significantly accelerate various open-source large language models across a diverse ecosystem of AI chips, including those from Nvidia, AMD, Google's TPU, Apple Metal، and Intel Arc.
ZML's core ambition, as articulated by founder Steeve Morin to TechCrunch, is to dismantle existing technological silos and unlock the full potential of disparate hardware for AI applications, often exceeding their stated maximum speeds. In an increasingly AI-driven world where inference (the processing of prompts) is gaining primacy over model training, the current landscape is often fragmented by software and architectural barriers, leading to vendor lock-in. ZML/LLMD offers a compelling solution, promising peak performance across heterogeneous chip arrays. This is not merely a technological triumph but a potential market disruptor, especially amidst growing concerns over AI-related operational costs. By enabling enterprises and cloud providers to leverage a mix of chips—potentially less costly or more energy-efficient alternatives—ZML aims to decentralize power and foster widespread, efficient AI dissemination.
Morin suggests that this software-driven assistance could be a boon for novel AI chipmakers, particularly those emerging from Europe, citing examples such as Axelera, Fractile, Kalray, OLIX, Q.ANT, SiPearl, SpiNNcloud, and VSORA. ZML's focus, however, transcends geographical origin; it’s about collaborating on "things that haven’t been done before anywhere in the world." Despite its disruptive potential, ZML isn't bearish on Nvidia; Morin emphasizes a positive relationship with the AI chip giant, acknowledging its strategic preparations for the surging demand in inference.
The "inference gold rush" has attracted intense investment and competition, with players like Baseten (recently valued at $13 billion), Inferact (from vLLM creators), and RadixArk (behind SGLang) also vying for market share. While vLLM and SGLang partially overlap with LLMD's offerings, Morin envisions a broader impact for ZML, stating, "We have reached the point where we are co-designing silicon." The startup's agility is attributed to its lean 20-person team in Paris, backed by significant funding for its size. Morin, with a successful track record as VP of engineering at Zenly (acquired by Snapchat for nine figures), secured $20 million from prominent venture firms including 20VC, >commit, AALVC, Drysdale Ventures, Kima Ventures, Kindred Capital, LocalGlobe, and Puzzle Ventures.
Unlike ZML's previous open-source ML framework, ZML/LLMD is a proprietary product launching as a free offering. This strategic decision aims to gather crucial usage data before establishing a revenue model. "I’d rather measure and [then generate revenue] where it is most effective without hindering my growth stupidly because I have been too greedy from the get-go," Morin explained. While its future monetization and adoption remain to be seen, ZML's cap table is a testament to its industry appeal, featuring tech luminaries such as Dagger and Docker founder Solomon Hykes, Hugging Face's Clément Delangue and Julien Chaumond, and Yann LeCun (now with AMI Labs). This underscores a broader narrative: Europe's AI startups are increasingly thriving and building innovation from their home turf, with Morin proudly asserting, "I couldn’t do ZML anywhere but in Paris."
المصدر: عرض الخبر الأصلي
مرحبا، أسمي زين وهذه مدونتي أسعى دائما لأقدم لكم أفضل المواضيع الخاصة بالتكنلوجيا
ليست هناك تعليقات:
إرسال تعليق