30 Days of AI - June 2024
Fabrice Bagniakana
EMEA GenAI Dell Business Development Manager I Sustainability : United Nations Decade on Ecosystem Restoration 2021-2030 (Task Force Member)
Dell launches Dell AI Factory with NVIDIA
The Dell AI Factory, introduced at Dell Technologies World 2024, addresses organizations' most pressing challenges with a portfolio of AI technologies, an open ecosystem of partners, validated and integrated solutions, and expert services to help achieve AI outcomes faster. This is the industry’s first end-to-end enterprise AI solution, designed to simplify and accelerate AI adoption within organizations. Notably, the Dell AI Factory with NVIDIA reduces setup time by up to 86% compared to doing it yourself.
Dell Technologies and NVIDIA have a long-standing partnership with over 25 years of joint innovation, focused on accelerating innovation and delivering cutting-edge platforms, solutions, and software that enable transformative results for joint customers. This ongoing partnership not only showcases a commitment to innovation but also demonstrates the ability to adapt to the evolving technological landscape and enable joint innovation across the entire portfolio—from client devices to servers and solutions, as well as data centers, cloud, and edge.
This approach enables customers to build their own AI factories from a broad set of capabilities right sized for their needs, offered as pre-validated, full stack solutions or as integrated components that are tailored to their needs to help them get started on AI use cases.
Meta unveils Meta Llama 3
Meta proudly unveils Meta Llama 3, a game-changing leap in open-source large language models. With models boasting 8 billion and 70 billion parameters respectively, Llama 3 not only excels across industry benchmarks but also introduces enhanced reasoning capabilities. Embracing openness and innovation, Meta places Llama 3 in the hands of the community, fostering collaboration and propelling AI advancements across domains. Committed to responsible AI, Meta equips users with tools like Llama Guard 2 and Code Shield, ensuring ethical usage. Looking forward, Meta plans to enhance Llama 3's capabilities further, promising multilingual support, longer context windows, and improved performance in core functionalities.
?
In the realm of AI, Meta AI, fueled by Llama 3, emerges as a leading assistant, facilitating productivity, creativity, and connectivity. With Meta Llama 3, the future unfolds with boundless opportunities for intelligent interactions and innovations, enriching human endeavors across the board.
You can try Meta AI?here.
*Please see?evaluation details?for setting and parameters with which these evaluations are calculated.
Running Meta Llama 3 Models on Dell PowerEdge XE9680
Dell Technologies recently conducted configuration tests with Llama 3, utilizing the PowerEdge XE9680 server equipped with 8x H100 GPUs, specially optimized for AI workloads. Leveraging the high-speed NVLink interconnect and 10 PCIe slots, the XE9680 proves ideal for deploying large models like Llama 3 70B, requiring multiple GPUs for optimal performance and memory capacity. In the tests, Dell deployed various Llama models, including Meta-Llama-3-8B and Meta-Llama-3-70B, alongside Llama 2 counterparts. These models were run with native precision, allowing for a fair comparison. Utilizing NVIDIA TensorRT-LLM as the inference framework and Rocky Linux 9.1 as the operating system, Dell focused on evaluating inferencing speed, specifically Time-to-First-Token (TTFT) and throughput across different batch sizes.
?
Given the substantial accuracy improvements of Llama 3, Dell concentrated on assessing inferencing speed, particularly emphasizing throughput over various batch sizes. Keeping input and output token lengths consistent across experiments, Dell observed notable improvements in throughput, especially with long input token lengths. Leveraging the capabilities of H100 GPUs to support the fp8 data format, Dell measured throughput under extended input token lengths for Llama 3 models, highlighting significant enhancements. This meticulous testing underscores the potential of Llama 3 to elevate AI inference performance, offering promising implications for various applications and industries.
?
NVIDIA Keynote News Wrap-Up
?NVIDIA CEO Jensen Huang delivered a powerful keynote ahead of COMPUTEX 2024, sharing the latest AI breakthroughs that are powering a new industrial revolution.
领英推荐
?See some of the many exciting announcements below and watch the replay for the full experience.
Watch the Keynote replay : NVIDIA at Computex 2024 | June 4-7, 2024 | NVIDIA
Mistral AI has launched Mixtral 8x22B
Mistral AI proudly introduces Mixtral 8x22B, a cutting-edge open model that redefines performance and efficiency standards in the AI domain. Leveraging a sparse Mixture-of-Experts (SMoE) architecture, Mixtral 8x22B utilizes a mere 39B active parameters out of 141B, offering unmatched cost efficiency without compromising on capabilities. This model boasts fluency in multiple languages including English, French, Italian, German, and Spanish, coupled with robust mathematics and coding prowess. With native function calling capabilities and a 64K tokens context window, Mixtral 8x22B facilitates precise information retrieval from large documents, empowering applications development and tech stack modernization at scale.
Google has launched Med-Gemini
Introducing Med-Gemini by Google, an advanced AI tailored specifically for medical applications.??
This groundbreaking AI surpasses previous models, including GPT-4, across a wide range of medical benchmarks. With an unprecedented accuracy of 91.1% on the challenging MedQA-USMLE benchmark, Med-Gemini sets a new standard in medical AI.??Outperforming GPT-4 by an impressive 44.5% on average across seven diverse benchmarks, Med-Gemini showcases its superior capabilities. Its proficiency extends to tasks such as medical summarization, generating doctor referrals, and simplifying complex medical documents. Med-Gemini emerges as the preferred choice over human expert analysis for intricate text-based medical tasks. This advancement represents a significant leap forward in AI for healthcare, promising potential enhancements in medical diagnostics and patient care. Positioning healthcare at the forefront of AI development, Med-Gemini holds the promise of substantial impacts on clinical practices and medical education.
Anthropic introduces Claude 3
Anthropic introduces Claude 3, a groundbreaking model family that redefines industry standards across cognitive tasks. Comprising Haiku, Sonnet, and Opus models, each offers escalating levels of performance, enabling users to tailor their selection based on intelligence, speed, and cost requirements. Opus, the pinnacle of intelligence in the family, surpasses peers in common AI evaluation benchmarks, exhibiting human-like comprehension and fluency in complex tasks.
Claude 3 models showcase enhanced capabilities in analysis, forecasting, content creation, code generation, and multilingual conversation. These models empower real-time applications like customer chats, auto-completions, and data extraction, where immediate responses are critical. Haiku stands out as the fastest and most cost-effective option, capable of processing information-rich research papers in record time. Sonnet, boasting twice the speed of its predecessors with heightened intelligence, excels in tasks necessitating rapid responses. Meanwhile, Opus maintains competitive speeds while offering unparalleled intelligence, signaling a significant leap forward in general intelligence modeling.
Microsoft launches Phi-3, its smallest AI model yet
Microsoft unveils Phi-3 Mini, the latest iteration of its lightweight AI model series and the first of three compact models in the pipeline. With 3.8 billion parameters, Phi-3 Mini offers a scaled-down alternative to larger language models like GPT-4, now accessible on Azure, Hugging Face, and Ollama platforms. Microsoft's roadmap includes forthcoming releases of Phi-3 Small (7B parameters) and Phi-3 Medium (14B parameters), marking a progressive expansion in model capabilities.
?
Following the success of Phi-2, which rivaled larger models like Llama 2, Microsoft's Phi-3 demonstrates enhanced performance compared to its predecessor. Despite its compact size, Phi-3 delivers responses akin to models ten times its size, underscoring Microsoft's commitment to developing efficient yet powerful AI solutions. With Phi-3 Mini paving the way, Microsoft continues to advance lightweight AI models, catering to diverse applications and accessibility needs.Microsoft unveils Phi-3 Mini, the latest iteration of its lightweight AI model series and the first of three compact models in the pipeline. With 3.8 billion parameters, Phi-3 Mini offers a scaled-down alternative to larger language models like GPT-4, now accessible on Azure, Hugging Face, and Ollama platforms. Microsoft's roadmap includes forthcoming releases of Phi-3 Small (7B parameters) and Phi-3 Medium (14B parameters), marking a progressive expansion in model capabilities.