Sam Poursoltan ,PhD

Azure has 1674 AI models in the portfolio (and counting), but more importantly, Azure has a range of benchmarks for these AI models. Take a look... very informative for accuracy, coherence, fluency, relevance, etc - https://lnkd.in/ge7HsBvV

7

1 Comment

Kolmogorov–Arnold Networks seem like a paradigm shift in ML architectures. Curious to see how they evolve. "While MLPs have fixed activation functions on nodes ("neurons"), KANs have learnable activation functions on edges ("weights"). KANs have no linear weights at all -- every weight parameter is replaced by a univariate function parametrized as a spline. " https://lnkd.in/gdPk_qCQ

29

How to fine-tune #LLMs on custom #datasets at Scale using Qwak and #CometML Alexandru Razvant https://lnkd.in/gP4aWM69

3

Meta has released Llama 3.2, which includes both vision and text models optimized for edge and mobile use: ->> Vision Models (11B & 90B): Drop-in replacements for text models, outperforming closed models like Claude 3 Haiku in image understanding. Fine-tuning is available with tools like torchtune and torchchat. ->> Lightweight Text Models (1B & 3B): Designed for mobile and edge devices, they support 128K tokens and excel in summarization, rewriting, and instruction tasks. Optimized for Qualcomm, MediaTek, and Arm processors. ->> Llama Stack Distributions: Simplifies deployment across environments (cloud, on-prem, on-device), supporting retrieval-augmented generation (RAG) and tooling-enabled apps with integrated safety features. ->> Ecosystem Support: Collaborations with AWS, Dell, Databricks, and others for enterprise clients. On-device distribution via PyTorch ExecuTorch and single-node via Ollama. ->> Open Availability: Llama 3.2 models can be downloaded from llama.com, Hugging Face, and partner platforms like AWS, Google Cloud, and Microsoft Azure. Meta emphasizes openness and modifiability to foster innovation. https://lnkd.in/gxCmurx8

6

✨ Big News in AI Development! ✨ Microsoft has rebranded Copilot Studio as Azure AI Foundry, giving us a better, more integrated platform for building and deploying AI solutions. Think seamless access to a huge model library, smarter workflows with Prompt Flow, and next-level information retrieval to make your copilots even more useful. Plus, you can take your data, pick a model, and have a fully functional chatbot in just 20 minutes. Azure AI Foundry is here to make the process of bringing AI projects to life smoother and faster. If you're into building innovative AI tools, this is something to watch. Let's see how it helps us create the next generation of smart tech! #Microsoft #AzureAIFoundry #AI #MachineLearning #TechInnovation

4

Who has never worked on relational database before? Relational database is no doubt the most popular database of all time. As popular as it is however, people always complain about how confusing and complex relational data modeling is. With the advent of NoSQL databases, this complaint seems to get some momentum, and now people are even questioning if data modeling is necessary at all. The truth is, yes, we need data modeling. On the other hand, the current data modeling practice is too relational-centric, causing all the questions and doubts and noises. We need a data modeling approach that is applicable to all real world scenarios and across different modeling techniques, either relational or NoSQL. In this blog series, I will introduce universal data modeling, which is a structured approach ensuring that every aspect of database design is grounded in a clear understanding of the underlying semantics, enabling seamless transitions between abstraction and implementation. Hope this series can help you in your data modeling practice. Cheers! #relational #datamodel #datamodeling Universal Data Modeling https://lnkd.in/g8FF9mzH

7

Interesting to see what data is actually going into training.

2

**Exciting News 🚀** We're thrilled to announce that Mosaic AI Model Training's support for fine-tuning GenAI models is now available in Public Preview 🤩 **Unlock the Power of Your Enterprise Data 🔓** At Databricks, we believe that connecting the intelligence in general-purpose LLMs to your enterprise data is the key to building high-quality GenAI systems. With Mosaic AI Model Training, you can fine-tune a wide range of models, including Llama 3, Mistral, DBRX, and more, with your enterprise data. This results in models that are specialized for specific tasks, business contexts, or domain knowledge. **Benefits of Fine-Tuning 📈** Our customers have seen significant benefits from fine-tuning, including: * **Higher quality**: Improve model quality along with specific tasks and capabilities * **Lower latency at lower costs**: Fine-tuning small models can dramatically reduce latency and cost while maintaining quality * **Consistent, structured formatting or style**: Generate outputs that follow a specific format or style * **Lightweight, manageable system prompts**: Integrate business logic or user feedback into the model itself * **Expand the knowledge base**: Extend a model's knowledge base with Continued Pretraining **Real-World Success Stories 📊** Don't just take our word for it Our customers have achieved amazing results with Mosaic AI Model Training: * Experian reduced average training time and accelerated their GenAI development cycle * First American automated tedious manual tasks and exceeded accuracy goals * Celebal Tech built a high-quality domain-specific RAG system with fine-tuning and RAG **Get Started Today 🚀** Try Mosaic AI Model Training today and discover the power of fine-tuning for yourself. With our simple, fast, and integrated solution, you can: * Select your base model and training dataset * Start training immediately * Deploy your model with Mosaic AI Model Serving in just one click **Learn More 📚** Watch our Data and AI Summit presentation, read our documentation, and visit our pricing page to learn more about Mosaic AI Model Training. #MosaicAIModelTraining #FineTuning #GenAI #Databricks #AI #MachineLearning #EnterpriseData #DataScience #ArtificialIntelligence #Innovation #Technology #BusinessSuccess #CustomerSuccess #AICommunity #DataCommunity #DatabricksCommunity

1

Pushing the edge on how to use #AI and #LLMs with data tables and (numerical) datasets to answer questions and help with more informed decisions. In this technical deep dive I go through how to build a fully scalable and LLM-friendly datastore to use in combination with knowledge systems. The context is #WASH data.

10

3 Comments

💡 How to develop foundation models in the graph domain? 🤖 Foundation models such as LLMs are dramatically changing the AI landscape due to their capacity to solve new tasks in a few-shot learning regime. A shared input space across all tasks, such as the vocabulary of tokens, is an essential ingredient through which LLMs manifest this ability in terms of generalization.

1

The complexity of enterprise data, which typically tends to be stored across multiple data tables, databases was largely ignored in Predictive AI till it could no longer be ignored. That realization brought #featureengineering, #featurestores and many other technologies front and center in Predictive AI. When it came to #generativeAI for tabular data, our team decided to take this head on, from the very beginning (dating back to 2016). This article by Neha Patki and Frances Hartwell from DataCebo talks about our evolution and how we continue to make progress on this front with The Synthetic Data Vault Read about it here: https://lnkd.in/eJSyArej #syntheticdata #generativeai

26

📢 Hey friends of Data Sciences and Machine Learning, I'm looking for a speaker for Perth Machine Learning Group to share ideas, learnings, experiments and knowledge for the August Meetup (In-person). DM me for more details. If you have a vague idea of what you might want to present, ping me anyway! I'll help you polish it! #AI #ML #LLM #SpeakingOpportunity

25

As many of you know, ETL and ELT have been central to my career in data for almost 30 years. We have seen many new developments in that time, but now it's getting more exciting—it turns out AI really can help us! One of the most time-consuming tasks in any data project is transforming and shaping the data for loading into the target schema. This is the T (the transformation) in ELT. 🎉 Exciting News! 🎉 Over the past months, Zachary H. has been developing "Cocoon" to automate this "T" within databases using Large Language Models (LLMs). It has been a thrill for me to guide him a little in this work. 🚀 We're thrilled to announce the preview release of the table staging feature today! Staging is often the first yet tedious step in data projects, involving tasks like deduplication, data type casting, handling missing values, correcting typos … and more.   Cocoon simplifies these tasks by automatically generating SQL code while allowing interactive user feedback. 🔗 Cocoon is open source: https://lnkd.in/gbdmPS2w   📊 Check out the demo notebook for staging tables in Snowflake/DuckDB, outputting DBT staging codes and YML files: https://lnkd.in/gcrM6Wkc We're excited to see how Cocoon transforms your data projects!

76

4 Comments

📢 Is your knowledge graph ready for 2025? 🧠 In the latest Practically Intelligent episode, Akshay and I sit down with Sudhir Hasbe, Chief Product Officer at Neo4j, to discuss why graph databases are revolutionizing Retrieval-Augmented Generation (RAG) and AI applications in general. Sudhir’s background as the product leader behind Google’s analytics products and his expertise in graph technology make this episode a must-listen if you’re even considering using a graph for your AI applications whether it's a RAG system, an AI agent, or something else! 🔍 We cover: - 🌟 Why 2024 was a breakout year for knowledge graphs - 🔗 How graph RAG integrates structured and unstructured data in a unified pattern - 📊 The trifecta of big data, cloud computing, and LLMs fueling graph adoption - 💡 Practical tips to implement graph RAG in your enterprise 🎧 Listen now: 👉 Apple: https://lnkd.in/gTxcUqit 👉 YouTube: https://lnkd.in/gywX5KjS Let us know in the comments: What’s your biggest takeaway from graph RAG, or how are you thinking about knowledge graphs in your work? 👇 #GraphDatabases #GraphRAG #KnowledgeGraphs #AI #PracticallyIntelligent #RAG #Agents

105

2 Comments

Want to get started with evaluating LLMs on Databricks? Check this video out ⏬

7

1 Comment