Databricks’ Post

There have been huge waves recently after #Databricks announced their new open-source fine-grained MoE (mixture of experts) large language model... but what does that actually mean? If you're already happily chatting away with your favorite LLM, why would you care? And how do you even get started with using it? In this video, Simon takes a look through the major points of the DBRX announcement and what the sheer focus on speed tells us about LLM maturity. We then take a dive into the new Databricks AI Playground to test out DBRX Instruct and show you how you get can get started with it. https://hubs.la/Q02trdVk0 #dataengineering #datalakehouse #datascience #machinelearning #llm #DBRX #genai

Super interesting reading through the announcements & technical explanations for #DBRX, and it's a much more compelling story than the original Dolly models, given the DBRX LLM performs pretty damn well! I'm no data scientist, but the engineering behind it is impressive to say the least! If you've not seen it before, there's a new "AI Playground" rolling out to Databricks workspaces, which supports simultaneously prompting multiple foundational AI models and measuring the results. If you're comparing & contrasting LLMs currently - it's worth checking it out! #datascience #machinelearning #AI #LLMs #GenAI #datalakehouse

We know having access to the newest LLMs is important to our users. That's why on the same day Databricks Mosaic Research launched their open source model DBRX, we added it our AI Model Selector ✨ You can try DBRX alongside all your other favorite AI models on You.com today! #youdotcom #dbrx

Is your enterprise stuck in the pilot phase of generative AI deployment?    Discover how LLMOps can propel you from experimentation to full-scale implementation, driving innovation and efficiency.    In this PoV, Ravindra Patil, Senior Director of Data Science at Tredence, explains how to capitalize on specialized processes to reduce operational costs, accelerate time to market, and increase productivity.    Don’t miss out on the transformative benefits of LLMOps to stay ahead in the competitive AI landscape. Download the POV https://lnkd.in/g2xgnGdX #GenAI #LLMOps #Tredence #BeyondPossible

Learn more about #DBRX at our AI Alliance meetup this Tuesday!

It was just 2 years ago that I led the first experimental deployment of AMD GPUs at Meta for AI workloads for both LLMs and Recommendation Systems culminating in the joint announcement by AMD and Meta about Meta's plan to deploy AMD GPUs https://lnkd.in/gGEvJ69z. Since then, a lot of work has happened to bring Accelerated Transformers and the scaled dot product attention (SDPA) operator to AMD GPUs, enable Triton for AMD GPUs and support AMD GPUs with PT2 compilation to make AMD GPUs fully competitive with the market leader Nvidia for training and inference of LLMs. I'm super excited about HuggingFace's contribution to enabling AMD GPUs for their ecosystem, as HuggingFace has been our trusted partner in LLM enablement for PyTorch from day 1. It's gratifying to see the strides AMD GPUs have made with availability on Microsoft's Azure clouds from initial PyTorch support in 2021 https://lnkd.in/gJMagJcW. Establishing a viable alternative to Nvidia increases competition -- and when companies compete with ideas and innovation, customers, engineers, our industry and solicitud at large win because products become more performant and cheaper. By growing the market rather than monopolizing marketshare, even the companies forced to compete ultimately win. A big shout-out to the AMD GPU enablement team for their work with many team members from my original enterprise AI team at IBM that pioneered software/hardware co-design for AI with PowerAI and the "Minsky" systems a decade ago https://lnkd.in/g_daV5zQ. #PyTorch #PT2 #AcceleratedAI #AcceleratedLLM #AcceleratedTranformers #ROCm #AMD #MetaAI #Meta

Fresh off of their summit this week, Databricks is popping up a lot in the AI narrative this week, and for good reason. Their grounded focus on working with enterprise data and use cases has them in a great position. One highlight is the demo of Mosaic AI done by Kasey Uhlenhuth. It was a bold swing to do a live demo but it pays off. Murphy's Law applies in great fashion to live demos, and as expected there is a glitch. Kasey's recovery in real time shows off the platform brilliantly, and the difference between leveraging the AI on true enterprise data versus open market is stark. Demo is halfway through the link. https://lnkd.in/e5Mi9uJp

Exciting news for CIOs: Snowflake's release of Polaris, an open-source Iceberg catalog, will enable businesses to access Iceberg data via Fivetran, unlocking rapid and secure value through AI and applications. Learn more about the partnership in our conversation with David Vellante and Rebecca Knight from SiliconANGLE & theCUBE. https://5tran.co/3Sx4lf2 #SnowflakeSummit24

Learn about #DBRX, a new general-purpose LLM, from the people who built it. The Databricks Mosaic AI family makes it production-ready to day via quality controls, model monitoring for drift & hallucinations, and guardrails to protect against your model training with PII or unsafe data. Databricks has the responsible, production-ready, end-to-end solution set for Enterprises today. DBRX outperforms all established open models in performance benchmarks and provides a reference design that you can use to quickly and efficiently build your own custom #LLMs. Join our webinar to learn how you can quickly get started with DBRX, build your own custom LLM, and more. Featuring Naveen Rao (Mosaic ML Founder and VP of Generative AI at Databricks) and Jonathan Frankle (Chief Scientist of Neural Networks).

Great working with visionary team at WWT and RackN to create the first fully automated AI platform vending machine. Pick your hardware config and OS distro... and everything else is done for you. No vendor lock-in, no stranded resources, no complicated fabric manipulation and full hardware/software configuration flexibility. Once you layer in your automated AI workflows you have the ability to flexibly tweak and tune and ultimately optimize any workload with the right set of servers, accelerators, networking, OS Distro, and AI framework and libraries. Accelerating AI time to market does not require you to cut corners in scenario testing and optimization. It requires you to test in a flexible and efficient manner.