Le projet THINK

To the Edge and Beyond’s Gabrielle Bejarano spoke with Intel Product Engineer Sean Williams on how Intel utilizes open-source solutions to overcome the communication hurdles and get industries the data, they need to solve pressing business problems. 

Intel’s Artificial Intelligence (AI) Solutions Group has generated USD 1.3B in business value so far and is exploring ways to scale AI across all of Intel’s business domains.

Today MLCommons® published industry results for their AI training v2.1 benchmark that contained an impressive number of submissions, with over 100 results from a wide array of industry suppliers, with roughly 90% of the entries for commercially available systems, ~10% in preview (not yet commercially available) and a few in “research,” not yet targeting commercial availability at this time. Habana® Gaudi®2 is one of three AI accelerators reporting results, with Nvidia A100 and H100 rounding out the category, with H100 in preview (not available yet).

Innovation 2022 AI Track: “Achieving efficient, productive and good quality End-to-End AI Pipelines”

One of the key challenges in Large Language Model (LLM) training is reducing the memory requirements needed for training without sacrificing compute/communication efficiency and model accuracy.  DeepSpeed [2] is a popular deep learning software library which facilitates memory-efficient training of large language models. DeepSpeed includes ZeRO (Zero Redundancy Optimizer), a memory-efficient approach for distributed training [5].  ZeRO has multiple stages of memory efficient optimizations, and   Habana’s SynapseAI® software currently supports ZeRO-1 and ZeRO-2. In this article, we will talk about what ZeRO is and how it is useful for training LLMs. We will provide a brief technical overview of ZeRO, covering ZeRO-1 and ZeRO-2 stages of memory optimization.  More details on DeepSpeed Support on Habana SynapseAI Software can be found at Habana DeepSpeed User Guide.  Now, let us dive into why we need memory efficient training for LLMs and how ZeRO can help achieve this.

Deploy Models Faster Best Practices: Learn about AI reference kits

Wei Li talks AI Everywhere with Ronald van Loon at World Summit AI

Every year people become lost or injured far from the trappings of civilization, and search and rescue can prove challenging due to harsh terrain and weather. That’s where leading-edge technology may help to turn the odds.

In this tutorial, we will demonstrate fine tuning a GPT2 model on Habana Gaudi AI processors using Hugging Face optimum-habana library with DeepSpeed.

Evaluating AI deployments and machine learning based on overall energy usage instead of just processing power is a new idea. It’s so new that there is no standard metric currently. Each section of the ML pipeline consumes an enormous amount of energy, and each section should be evaluated and enhanced.