Researchers from UC Santa Cruz, UC Davis, LuxiTech, and Soochow University have developed a new method to run AI language models more efficiently by eliminating matrix multiplication, potentially reducing the environmental impact and operational costs of AI systems. Ars Technica's Benj Edwards reports: Matrix multiplication (often abbreviated to "MatMul") is at the center of most neural network computational tasks today, and GPUs are particularly good at executing the math quickly because they can perform large numbers of multiplication operations in parallel. [...] In the new paper, titled "Scalable MatMul-free Language Modeling," the researchers describe creating a custom 2.7 billion parameter model without using MatMul that features similar performance to conventional large language models (LLMs). They also demonstrate running a 1.3 billion parameter model at 23.8 tokens per second on a GPU that was accelerated by a custom-programmed FPGA chip that uses about 13 watts of power (not counting the GPU's power draw). The implication is that a more efficient FPGA "paves the way for the development of more efficient and hardware-friendly architectures," they write.
The paper doesn't provide power estimates for conventional LLMs, but this post from UC Santa Cruz estimates about 700 watts for a conventional model. However, in our experience, you can run a 2.7B parameter version of Llama 2 competently on a home PC with an RTX 3060 (that uses about 200 watts peak) powered by a 500-watt power supply. So, if you could theoretically completely run an LLM in only 13 watts on an FPGA (without a GPU), that would be a 38-fold decrease in power usage. The technique has not yet been peer-reviewed, but the researchers -- Rui-Jie Zhu, Yu Zhang, Ethan Sifferman, Tyler Sheaves, Yiqiao Wang, Dustin Richmond, Peng Zhou, and Jason Eshraghian -- claim that their work challenges the prevailing paradigm that matrix multiplication operations are indispensable for building high-performing language models. They argue that their approach could make large language models more accessible, efficient, and sustainable, particularly for deployment on resource-constrained hardware like smartphones. [...]
The researchers say that scaling laws observed in their experiments suggest that the MatMul-free LM may also outperform traditional LLMs at very large scales. The researchers project that their approach could theoretically intersect with and surpass the performance of standard LLMs at scales around 10^23 FLOPS, which is roughly equivalent to the training compute required for models like Meta's Llama-3 8B or Llama-2 70B. However, the authors note that their work has limitations. The MatMul-free LM has not been tested on extremely large-scale models (e.g., 100 billion-plus parameters) due to computational constraints. They call for institutions with larger resources to invest in scaling up and further developing this lightweight approach to language modeling.
MTVNews.com has been shut down, with more than two decades' worth of content no longer available. "Content on its sister site, CMT.com, seems to have met a similar fate," adds Variety. From the report: In 2023, MTV News was shuttered amid the financial woes of parent company Paramount Global. As of Monday, trying to access MTV News articles on mtvnews.com or mtv.com/news resulted in visitors being redirected to the main MTV website.
The now-unavailable content includes decades of music journalism comprising thousands of articles and interviews with countless major artists, dating back to the site's launch in 1996. Perhaps the most significant loss is MTV News' vast hip-hop-related archives, particularly its weekly "Mixtape Monday" column, which ran for nearly a decade in the 2000s and 2010s and featured interviews, reviews and more with many artists, producers and others early in their careers. "So, mtvnews.com no longer exists. Eight years of my life are gone without a trace," Patrick Hosken, former music editor for MTV News, wrote on X. "All because it didn't fit some executives' bottom lines. Infuriating is too small a word."
"sickening (derogatory) to see the entire @mtvnews archive wiped from the internet," Crystal Bell, culture editor at Mashable and one-time entertainment director of MTV News, posted on X."decades of music history gone... including some very early k-pop stories."
"This is disgraceful. They've completely wiped the MTV News archive," longtime Rolling Stone senior writer Brian Hiatt commented. "Decades of pop culture history research material gone, and why?"
The report notes that some MTV News articles may be available via internet archiving services like the Wayback Machine. However, older articles aren't available.
An anonymous reader quotes a report from Reuters: Alphabet's Waymo said on Tuesday its autonomous ride-hailing service, Waymo One, is now available to everyone in San Francisco, nearly four years after a similar move in Phoenix, Arizona. Driverless vehicles are expected to drive commercial success for automakers even as regulatory scrutiny remains tight amid concerns of investors about growing investments in the nascent technology. Waymo had started a test service with its research-focused program in San Francisco in 2021, which included an autonomous specialist on board for all rides at that time, as it looked to commercialize the technology.
The company said that about 300,000 people had signed up to ride with Waymo since it first opened a waitlist in the city, signaling strong demand. Now with open access, anyone can request a ride on its app. The company had opened access to everyone in Phoenix, Arizona without a waitlist in 2020. Mountain View, California-based Waymo is a self-driving technology pioneer, which started its first U.S. driverless taxi service in 2020 over a decade after it was born in 2009 as a project inside Google. In March, the company received approval from the California Public Utilities Commission (CPUC) to start its Waymo One in Los Angeles and some cities near San Francisco.
Freiné par les ratés de sa chaîne d’approvisionnement, l’avionneur européen a annoncé lundi qu’il ne réussirait pas à livrer 800 appareils en 2024, mais plutôt 770. La Bourse l’a fortement sanctionné.
A l’issue d’un procès commencé en 2015, huit juges sur onze ont décidé que l’usage personnel ne ferait plus l’objet de sanction mais resterait un acte illicite. Il ne s’agit « en aucun cas d’une légalisation », a précisé le président de la plus haute juridiction.
Anglais et Slovènes peuvent se satisfaire de ce score nul et vierge (0-0). Les premiers restent invaincus et finissent premiers de leur groupe. Les deuxièmes se qualifient pour les huitièmes de finale pour la première fois de leur histoire.
Connexion
