Apple’s groundbreaking AI research could transform the iPhone and leapfrog Google and Microsoft | Technology News

But things are about to change. In a new research paper, Apple has demonstrated breakthrough technology that helps run AI on the iPhone. This technique involves streamlining bulky LLMs using flash storage optimization. It will also be a big deal when Apple integrates advanced AI into the iPhone. The Cupertino-based tech giant announced significant developments in AI through his two new research papers published this month. This paper reveals new techniques for 3D avatars and efficient language model inference.

The new study, “LLM in a Flash: Efficient Large Language Model Inference with Limited Memory,” published on December 12th, can provide a more immersive visual experience and allow users to access complex AI systems on their iPhone. , has the potential to transform your iPhone experience. iPhone and iPad. Research papers primarily focus on efficiently running large language models on devices with limited DRAM capacity. DRAM is a dynamic random access memory used in PCs and is known for its high speed, high density, affordability, and low power consumption.

Here are some takeaways from the research that put Apple ahead of its peers.

In this paper, we address the challenge of running LLM beyond the practically available DRAM, storing model parameters in flash memory, and executing them on demand to DRAM. We describe an inference cost model developed to optimize data transfer from flash memory, considering the characteristics of flash and DRAM.

The techniques described in this paper include windowing, which reduces data transfer by reusing previously activated neurons, and matrix banding, which increases data chunk size for efficient flash memory reads. It’s a ring.

This paper also focuses on exploiting sparsity in feedforward network (FFN) layers to selectively load parameters to increase efficiency. Another important aspect is memory management, which proposes strategies to efficiently manage the data loaded into DRAM and minimize overhead.

The researchers demonstrated their approach using models such as OPT 6.7B and Falcon 7B. According to the paper, the results showed that the model achieved 4-5x and 20-25x speedup on CPU and GPU, respectively, compared to traditional methods.

Regarding the practical application of the research, the two models demonstrated significant improvements in resource-limited environments.

New research from Apple presents an innovative approach to efficiently run LLM in hardware-constrained environments. This opens new directions for future research in on-device and next-generation user experiences.

What does that mean for iPhone users?

From a user perspective, discoveries about efficient LLM inference with limited memory could greatly benefit both Apple and iPhone users. Powerful LLM runs efficiently on devices with limited DRAM, such as iPhones and iPads, allowing users to immediately experience enhanced AI capabilities. These features include improved language processing, a more sophisticated voice assistant, increased privacy, the potential to reduce internet bandwidth usage, and most importantly, every iPhone user has access to advanced AI. This includes being able to access and respond.

Regardless of Apple’s future ability to demonstrate how it is working towards dominating AI research and applications, experts appear to be in alert mode. Some experts seem to suggest that tech giants need to take great care and responsibility when incorporating research results into real-world use cases. Some emphasize the need to consider privacy protection, ways to mitigate potential abuse and overall impact.

© IE Online Media Services Pvt Ltd

Bijin Jose is a technology journalist currently working as an assistant editor at Indian Express Online in New Delhi. Passionate about the transformative power of technology, he delves into the multidimensional impact technology has on our lives and the wider world. His interest in the vast possibilities of technology began at a young age and set the stage for his career. His writing career began at Maharaja Sayajirao University in Vadodara, where he earned a bachelor’s degree in English. At the same time, he took on the role of citizen journalist with Times of India to shed light on civic issues in Vadodara. After he completed his master’s degree in English Literature, he moved to the position of senior correspondent at TOI, focusing on crime reporting in print, while also producing features on the city’s culture, arts and heritage. In his 2016 he embraced the digital realm with India Today Digital. Starting out as a desk staffer, he quickly rose through the ranks to lead his team at Online News. His versatility shone through when he took on a mixed media role in the Times of India’s New Delhi bureau, organizing interviews with diplomats and defense officials and covering important events for the Response editorial department. . A passionate film buff, he writes film reviews and featurettes and contributes to The Economic, The Times and The Daily O. In 2021, his focus shifted to economics and technology journalism for the Times. By 2023, he had found his niche with his The Indian Express, where he continued to produce insightful features on the ever-evolving world of technology. His reports and stories are characterized by his talent for simplifying complex concepts and enlightening readers with the latest developments. … read more