elie.de

IBM 💙 Open Source Our AI platform, watsonx, is powered by a stack of open source tech, enhancing AI workflows enterprise readiness. Here's the list of key projects: 𝗧𝗿𝗮𝗶𝗻𝗶𝗻𝗴 & 𝗩𝗮𝗹𝗶𝗱𝗮𝘁𝗶𝗼𝗻: - CodeFlare: Simplifies the scaling and management of distributed AI workloads by providing an easy-to-use interface for resource allocation, job submission, and workload management. - Ray / KubeRay: A framework for scaling distributed Python workloads. KubeRay integrates Ray with Kubernetes, enabling distributed AI tasks to run efficiently across clusters. - PyTorch: An open-source framework for deep learning model development, supporting both small and large distributed training, ideal for building AI models with over 10 billion parameters. - Kubeflow Training Operator: Orchestrates distributed training jobs across Kubernetes, supporting popular ML frameworks like PyTorch and TensorFlow for scalable AI model training. - Job Scheduler (Kueue/MCAD): Manages job scheduling and resource quotas, ensuring that distributed AI workloads are only started when sufficient resources are available. 𝗧𝘂𝗻𝗶𝗻𝗴 & 𝗜𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲: - KServe: A Kubernetes-based platform for serving machine learning models at scale, providing production-level model inference for frameworks. - fms-hf-tuning: A collection of recipes for fine-tuning Hugging Face models using PyTorch’s distributed APIs, optimized for performance and scalability. - vLLM: A fast and flexible library designed for serving LLMs in both batch and real-time scenarios. - TGIS (Text Generation Inference Server): IBM’s fork of Hugging Face’s TGI, optimized for serving LLMs with high performance. - PyTorch: Used for both training and inference, this is a core framework in watsonx. - Hugging Face libraries: Offers a rich collection of pre-trained models and datasets, to provide cutting-edge AI capabilities. - Kubernetes DRA/InstaSlice: DRA allows for dynamic resource allocation in Kubernetes clusters, while InstaSlice facilitates resource sharing, particularly for GPU-intensive AI tasks. 𝗔𝗜 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁 𝗟𝗶𝗳𝗲𝗰𝘆𝗰𝗹𝗲: - Kubeflow & Pipelines: Provides end-to-end orchestration for AI workflows, automating everything from data preprocessing to model deployment and monitoring. - Open Data Hub: A comprehensive platform of tools for the entire AI lifecycle, from model development to deployment. - InstructLab: A project for shaping LLMs, allowing developers to enhance model capabilities by contributing skills and knowledge. - Granite models: IBM’s open source LLMs, spanning various modalities and trained on high-quality data. We're committed to the future of Open Source and its impact on the AI community.

🚀 Wir haben den Smart Country Startup Award 2024 im Bereich GovTech gewonnen! 🚀 Letzte Woche hatten wir das große Privileg, auf der SmartCountryConvention in Berlin für unsere innovativen Lösungen zur Digitalisierung des öffentlichen Sektors ausgezeichnet zu werden. 🏆 Wir setzten uns dafür ein, Städte und Verwaltungen mit modernster KI-Technologie zu unterstützen und effizienter zu gestalten. Dieser Award ist eine Bestätigung unserer Vision, durch intelligente Systeme die Arbeitsweise des öffentlichen Sektors zu revolutionieren und Deutschland im globalen KI-Wettlauf nach vorne zu bringen. 🌍💡 Ein großer Dank geht an unser engagiertes Team, unsere Partner und alle, die uns auf dieser Reise unterstützen. Diese Auszeichnung bestärkt uns, weiterhin innovative und ethisch verantwortungsvolle Lösungen für den digitalen Wandel zu entwickeln. 💛 #Govtech #Digitalisierung #KI #SmartCity #PublicSector #StartupAward #Innovation

One of the best visualization out there. Convolutional Neural Networks (CNNs) in action. ↓ Check out https://AlphaSignal.ai to get a daily summary of breakthrough models, repos, and papers in AI. Read by 180,000+ engineers and researchers.

It gives me a lot of comfort knowing that we are the last generation without advanced robots everywhere. Our children will grow up as “robot natives”. They will have humanoids cook Michelin dinner, robot teddy bears tell bedtime stories, and FSD drive them to school. We are the generation of “robot immigrants”, en route to a new world of ubiquitous Physical AI, much like our parents are “digital immigrants”, learning to realign their lives on 6 inches of touch screen. It’s a journey of both inventing sci-fi tech and reinventing ourselves. Everything that moves will be autonomous. Every year from now on will be the Year of Robotics. Here’s to a wild 2025 ahead 🥂

This is really something. The visuals are obviously AI generated but super interesting and illustrate consistent choices by the human artist. I was surprised, though, that the song was also entirely AI generated - I quite liked it and would not have guessed. Created by David Ljungberg who, does it all himself in his words: “I generate the song with Suno, create images on MidJourney, upscale them with Magnific, animate them using KlingAI, and finish editing in CapCut.”

Many 𝘈𝘐 𝘢𝘨𝘦𝘯𝘵𝘴 shared on LinkedIn are 𝘈𝘐 𝘸𝘰𝘳𝘬𝘧𝘭𝘰𝘸𝘴 or 𝘢𝘶𝘵𝘰𝘮𝘢𝘵𝘪𝘰𝘯𝘴 in disguise. Here's why this distinction 𝘮𝘢𝘵𝘵𝘦𝘳𝘴: AI agents are the current buzzword on LinkedIn. I won't complain about it: this puts 9x in the spotlight - and it's great to see countless posts showcasing examples of AI agents. Except... most posts I see aren't showing actual agents. Very often, these are AI automations. Or even, sometimes, "regular" automations. What's the difference? → 𝘈𝘶𝘵𝘰𝘮𝘢𝘵𝘪𝘰𝘯𝘴 execute predefined, rule-based tasks automatically. → 𝘈𝘐 𝘸𝘰𝘳𝘬𝘧𝘭𝘰𝘸𝘴 are automations that call LLMs like ChatGPT via API for one or more steps. → 𝘈𝘐 𝘢𝘨𝘦𝘯𝘵𝘴 are programs designed to perform non-deterministic tasks autonomously. Which tasks can they best handle? → 𝘈𝘶𝘵𝘰𝘮𝘢𝘵𝘪𝘰𝘯𝘴 shine with pre-defined deterministic tasks → 𝘈𝘐 𝘸𝘰𝘳𝘬𝘧𝘭𝘰𝘸𝘴 are great for deterministic tasks requiring some flexibility → 𝘈𝘐 𝘢𝘨𝘦𝘯𝘵𝘴 should be used to handle non-deterministic, adaptive tasks What are their strengths? → 𝘈𝘶𝘵𝘰𝘮𝘢𝘵𝘪𝘰𝘯𝘴 deliver outcomes reliably and are fast to execute → 𝘈𝘐 𝘸𝘰𝘳𝘬𝘧𝘭𝘰𝘸𝘴 are great for pattern recognition and handling complex rules → 𝘈𝘐 𝘢𝘨𝘦𝘯𝘵𝘴 are best when you expect new variables and scenarios What are their respective weaknesses? → 𝘈𝘶𝘵𝘰𝘮𝘢𝘵𝘪𝘰𝘯𝘴 limited to tasks explicitly programmed and cannot adapt → 𝘈𝘐 𝘸𝘰𝘳𝘬𝘧𝘭𝘰𝘸𝘴 require data to train models and are usually harder to debug → 𝘈𝘐 𝘢𝘨𝘦𝘯𝘵𝘴 are less reliable and may produce unpredictable outcomes --- Businesses want AI agents now - but many actually need good old automation. Understand the difference, and remember that what matters in the first place is solving problems with technology, no matter which one ends up being used in the end.

Reimagine your business from the ground up. Companies that prioritize customer needs and adopt a cross-functional, tech-enabled approach are achieving breakthrough results—like 80% faster operations and billions in new profits. Discover the benefits of putting customers at the center➡️hthttps://lnkd.in/gmD2t_Ea

Greet #Memary 🙂

A late xmas gift from me to you :-) take your baby-models to Model University for free! I've been training desktop models (1B Llamas) this holiday break, I wrote this tool to help train a desktop tax advisor (860 questions, 20 minutes training, wicked good 23%->61% accuracy for 1B compared to 97% of 70B) - here you go: Using Llama 3.3 and Groq - modeluni creates a curriculum of topics to teach and test the baby-model on; then for each topic/subtopic it creates varying levels of questions for training, and multi-choice questions for testing (that avoids seen questions). Baby-model ready for training with Unsloth AI on colab :-) and deployable back thru hugginface/Ollama. Then using Comet's Opik modeluni evaluates the model against the multi-choice questions, benchmarking against 3.3 and untrained baby-model. All configurable in a simple config file. Code here: https://lnkd.in/gGSaQNbW Enjoy -- feedback, ideas for improvement more than welcome :-) P.s, if you want to play with the tax advisor: https://lnkd.in/ggQB66E6 Tax advisor exam: https://lnkd.in/gw4jcGxV

Is your business still held back by tech from the '90s? 🖥️ Gen AI is slashing modernization timelines and unlocking value hidden in legacy systems. The catch? It’s all about how you build and scale. Explore the next-gen IT playbook today. https://mck.co/4ivY3rz