Sumitra Maharjan’s Post

I’m incredibly excited to share a milestone for TensorOps: our LLMstudio Proxy Server is now live on the AWS Marketplace! As the CTO of the company, I’m constantly thinking about how to make cutting-edge AI tools more accessible and easier to integrate. The LLMstudio Proxy Server is a big step in that direction. Whether you're building applications around large language models or scaling your existing AI systems, this tool is designed to give you the flexibility and control you need. Here’s what we’re offering: 🔹 No-Cost Tier: Deploy and manage the proxy server yourself. 🔹 Tiered Support: Our team will work with you to integrate LLMstudio into your apps, while also prioritizing your feature requests for the proxy server. This launch is a reflection of our commitment to empowering teams with scalable, easy-to-use solutions that unlock the full potential of AI. Take a look at our offering on AWS Marketplace: https://lnkd.in/d4DdwSnm I’d love to hear your thoughts, feedback, or even how you might use LLMstudio in your own projects. Let’s keep pushing the boundaries of what’s possible with AI together! #AWSMarketplace #LLMstudio #AIIntegration #ProxyServer #TensorOps #Innovation

Exciting times ahead in the world of AI and large code bases! The concept of Artificial General Intelligence (AGI) is both fascinating and daunting. Imagine having the ability to ask AGI to create a new version of Google in just a few hours and then hosting it at scale on Azure! Listening to Elon or Jensen and you might think we are just months or just a few years away from this being possible. Watching the amazing pace and results of SORA and other magical leaps in AI, perhaps it does feel possible. After all, isn't a video showing fluid dynamics created via GenAI proof that it can do anything? I don't believe we've reached that point yet, even with V1 AIG. One challenge is that for every action taken to impact the digital or physical world comes with consequences, as we know from the law: For every action, there is an equal and opposite reaction. The complexity of code bases like Google's search algorithms or Microsoft's Azure services highlights the intricate nature of today's systems. With architectures that have remained relatively unchanged over the last decade, a new approach is needed to unleash AI's full potential. During my tenure as a partner at Microsoft, I became acutely aware of how outdated architectural models were clung to by established engineers, significantly hindering the speed and quality of progress. It is clear that traditional monolithic, thread-based, container/host layer, and REST-based client/server architectures were designed to help humans tackle immediate challenges. However, a new, more flexible, dynamic, and semantic layer will abstract these rigid systems, enabling a new era of AI solutions. This future will be shaped by AI, not just as a designer but as its own operating system, capable of growing and adapting to any challenge. This cannot be achieved using decades-old concepts. Recognizing this, I have funded and assembled my own development team to explore innovative solutions to address the 'attention' issues that will hold AI back from creating large systems. The early results have exceeded my expectations, and the journey ahead promises groundbreaking advancements. #AI #ArtificialIntelligence #CodeBases #TechInnovation #MoreThanAttention #BraveNewThoughts

𝐃𝐞𝐞𝐩𝐒𝐞𝐞𝐤-𝐕𝟑: 𝐀𝐈 𝐉𝐮𝐬𝐭 𝐋𝐞𝐯𝐞𝐥𝐞𝐝 𝐔𝐩! Hey everyone, get ready for something amazing! For those new to it, DeepSeek is a cutting-edge, open-source large language model (LLM) project similar to what companies like Google or OpenAI are doing. It is known for its great performance and efficiency. Today, I'm thrilled to share the arrival of their latest version: DeepSeek-V3! It feels like just yesterday I was experimenting with V2, and now, BAM! We're seeing a massive jump. 𝗛𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝗺𝗮𝗸𝗲𝘀 𝗗𝗲𝗲𝗽𝗦𝗲𝗲𝗸-𝗩𝟯 𝗴𝗿𝗼𝘂𝗻𝗱𝗯𝗿𝗲𝗮𝗸𝗶𝗻𝗴: • 𝐁𝐥𝐚𝐳𝐢𝐧𝐠 𝐒𝐩𝐞𝐞𝐝: 60 tokens per second - that's 3x faster than V2! This means more efficiency and tackling bigger tasks becomes a breeze. I remember the early days of AI, and this progress is simply mind-blowing. • 𝗘𝗻𝗵𝗮𝗻𝗰𝗲𝗱 𝗖𝗮𝗽𝗮𝗯𝗶𝗹𝗶𝘁𝗶𝗲𝘀: DeepSeek-V3 isn't just faster; it's smarter. The team has significantly improved its capabilities, opening doors to even more innovative applications. • 𝗦𝗲𝗮𝗺𝗹𝗲𝘀𝘀 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻: API compatibility is intact! This means a smooth transition for developers - a true dream come true from a developer's point of view. • 𝗙𝘂𝗹𝗹𝘆 𝗢𝗽𝗲𝗻-𝗦𝗼𝘂𝗿𝗰𝗲: Sticking to their principles, DeepSeek has made both the models and research papers fully available. This fosters collaboration and pushes the entire AI field forward. I love seeing this kind of transparency in the industry. I'm incredibly excited to see what the community builds with DeepSeek-V3. The possibilities feel limitless. #DeepSeek #AI #ArtificialIntelligence #MachineLearning #OpenSource #Innovation #Technology #LLM

🚀 New Additions for the Modern Tech Stack Building for the future means adopting the latest tools and paradigms. Alongside your core primitives, consider these: 🔹 Vector Database – Essential for embedding storage and fast similarity searches. Powers workflows like RAG (Retrieval-Augmented Generation), semantic search, and recommendations. 🔹 AI Model (LLM) – Large Language Models are the brain behind next-gen features like personalization and generative experiences. Hosted APIs or fine-tuned models integrate seamlessly. 🔹 Edge Compute – The future is here: runtime and CDN unified into one. Platforms like Cloudflare Workers and Vercel Edge bring compute closer to the user for unparalleled speed. The stack is evolving—time to upgrade yours? 💡 #AI #EdgeComputing #TechStack

A new minion in my stack - Fabric I've added Fabric - an opensource project by daniel messler - to my AI toolkit, and it's really intriguing. It is certainly not mainstream and somehow doubt that it goes that way, but its open-source framework offers a unique and forward-thinking approach to handling AI tasks. The crowd maintained patterns - aka promps and its ability to pipe specific tasks together, connecting to different models is as "agentic" as I see reliable solutions right now - without the independence though. But lets wait. #AI #OpenSource #Innovation #TechTools

Day 1 Highlights: 5-Day Gen AI Intensive by Google & Kaggle Task1 The journey into Generative AI started on a strong note. We kicked off by diving into foundational concepts through listening to podcasts and reading whitepapers on large language models and prompt engineering. These resources laid the groundwork for understanding how LLMs operate and their role in text generation. On the practical side, I worked on a Kaggle codelab using the Gemini API, exploring its features and experimenting with different Google models. It was a great way to see how prompt techniques and parameter adjustments can shape AI interactions. We’ll also be following daily live streams with Google engineers, covering various topics and gaining insights directly from the experts. Excited to see how things unfold in the coming days. Stay tuned! #GenerativeAI #PromptEngineering #GoogleKaggle #LLM #AIInnovation

Facebook & instagram data are here

MLflow has emerged as a vital tool for optimizing LLM deployments, especially in the face of rapidly evolving AI solutions. In this video, Miloš Švaňa from profiq demonstrates how MLflow's latest features simplify the comparison and deployment of large language models. From setting up deployment servers to fine-tuning and prompt engineering, this hands-on guide is helpful for navigating the AI landscape. Watch the video here: https://lnkd.in/g9Z8m8dV #MLflow #LLMdeployment