io.net’s Post

Join @Supermicro and #Lightbitslabs on July 30th to learn how to keep your AI models humming at full speed with a data platform that delivers the lowest latencies, highest performance, and scalability, all while lowering your infrastructure costs. Find out how Lightbits' high-performance, scalable block storage and Supermicro's powerful computing solutions can transform your infrastructure to optimize your AI workflows. Register now! https://ow.ly/pv5N50Sy5UZ #AI #AIworkflows #Blockstorage #AImodels #dataplatform Sagy Volkov Janet Lafleur

Join Supermicro and Lightbits Labs for a panel discussion on AI Model Training and how Lightbits can accelerate your applications. Don't miss this great discussion. #AICloud #LLM #AI #storage #softwaredefinedstorage

With the soaring cost of data center GPUs and a growing amount of inefficiencies across the entire infrastructure landscape, optimizing your GPU/AI infrastructure investment is more important than ever. Take a look at some of the results Oak Ridge National Labs has seen with ArcHPC. DRASTICALLY increase your GPU performance WITHOUT buying more GPUs!

Every AI / LLM workload requires an incredible amount of compute and we are creating a compute economy. Exabits has incredibly scarce enterprise-grade GPU's like the famed H100s & A100s. We are building a subnet protocol using our accelerated, stabilized compute as the root. Literally, anyone can participate and be active in this economy! We are deeply committed to democratizing AI and this is an essential step. 👇👇👇👇👇👇👇👇👇👇👇👇 Join the new #web3 revolution today! #supply: exabits.ai/supply #discord: discord.gg/exabits #web: exabits.ai #telegram: t.me/exabitsofficial

The foundation for the AI stack is the infrastructure consisting of scalable compute and ubiquitous connectivity. Enabling AI infrastructure requires focused attention on open, scalable and power efficient solutions. This is the Pilot Episode of our "Enabling AI" Video Series. #EnablingAI #ConnectedbyBroadcom

Sebastian Barros of Ericsson is on to something good - a standard #Telecom #LLM! I agree wholeheartedly. If Ericsson and other industry leaders are support this, EnterpriseWeb is ready to contribute. Since leading the first ETSI #NFV poc, we've demonstrated how a harmonized standards-based telecom knowledge graph can enable real-time intelligent automation across the RAN, Core and Transport, and between layers 1-7 (in 50mb!). Last year we won an award for "Telco-grade generative AI for intent-based orchestration" in collaboration with Microsoft - https://lnkd.in/gJCUeHfq. We have trained small Open Source models for on-premise and edge use-cases. For #MWC24 we have two #GenAI / #AIOps demos. One for Intel Software - https://lnkd.in/eZiC6RGN and one with Red Hat https://lnkd.in/eBEWR87w. It's time to abstract complexity to enable and accelerate Telecom transformation initiatives (for real this time).

30x speed improvement for AI model training. Not April Fools. Keep in mind the below refers to pretraining LLama2-70B from scratch, not just fine-tuning it. Cerebras Hardware Architecture: https://lnkd.in/gdUTrTnt Cerebras ML Deep Dive: https://lnkd.in/gqituB4h

When we say the word “#latency,” most people have a clear definition in mind: the delay that occurs while a system waits for a packet to finish moving over a network. ⏱ This definition is technically correct, but incomplete. Fact is : there are multiple types of latency, and this definition only accounts for one of them - #network #latency. 🕸 So how do we define #compute #latency ?? 💻 And why is it so relevant to customers who work on their own #AI strategies ???

We need our own Telco Large Language model (s) ! General large language model will not be enough for the specific needs Telco industry will demand from AI in future. We need to act fast. As an inspiring case, the BloombergGPT model stands as a testament to the power of industry-specific Large Language Models (LLMs). With 50 billion parameters trained on a hybrid dataset of over 700 billion tokens, BloombergGPT has redefined excellence in the financial domain, trained using a AWS's robust computing infrastructure, including 64 p4d.24xlarge instances equipped with NVIDIA A100 GPUs. This breakthrough not only showcases the model's superior performance on financial benchmarks but also serves as a strong call to action for the telecommunications industry. With an annual capex exceeding $300 billion, the sector is perfectly positioned to invest in a Telco-specific LLM. Such a model could revolutionize network design, planning, optimization, and software development, harnessing the industry's vast datasets and technological skills. Moreover, the telecommunications industry's commitment to collaboration and standardization, exemplified by bodies like GSMA and TM Forum, provides a fertile ground for developing a unified, powerful LLM. This initiative would not only propel the industry forward but also set a new standard for innovation and collaboration. What few months seemed like a distant dream, or a utopian challenge is now within reach, thanks to rapid advancements and collaboration in training data, tokenization, model architecture, hardware development and refinement processes. BloombergGPT is a living example of the path telecommunications sector should embrace in AI with specialized LLMs. It's time for us to leverage our collective resources, expertise, and data to build an AI that not only understands our industry but also reshapes its future. https://lnkd.in/geqVAzPC #AI #LLM #telco #bloomberg #gsma #tmforum #bloombergGPT