Stephen Batifol

Deep Dive into xLSTM with Max Beck at ML Week Europe Join us at Machine Learning Week Europe for a keynote by Maximilian Beck from Johannes Kepler Universität Linz and our NXAI team. Max will talk about “xLSTM: New Architectures for Large Language Models”, sharing his latest breakthroughs in efficient and scalable AI models. 👉 Ready to explore the future of AI? Join Max and the NXAI team in Munich at Machine Learning Week! #MachineLearning #xLSTM #AI #MLWeekEurope #NXAI Peter Seeberg, Martin Szugat

111

4 Kommentare

The NeurIPS Navigator: A Behind the Scenes Look, Powered by Cerebras Inference ⚡ How do you create RAG app that turns 4,500+ research papers from the 2024 NeurIPS Conference? Dive into how we processed PDFs, created embedding chunks and stored it in on Postgres in a performant way. Created by our Fellow Isaac Tai Read more here: https://bit.ly/4fUOf8m

59

2 Kommentare

🚀 Looking to streamline fine-tuning for models like Llama 3, Mistral, and Mixtral? Our latest blog shows how Anyscale simplifies LLMs fine-tuning with Ray: 🔑 Key takeaways: • Scalable fine-tuning with LLM-Forge • Model Deployment with RayLLM Read more: https://lnkd.in/gQwD4A2U

35

The Rails 8 feature set perfectly complements SQLite's power in creating resilient, high-performance production apps, but still the question lingers: Can I really go all-in on #SQLite? Join Stephen Margheim at #RailsWorld as he illustrates how to leverage Rails and SQLite's full potential in your next venture, and when SQLite does and doesn't make sense for your application. https://lnkd.in/eaBstScy

52

1 Kommentar

🚀 Just slashed our ML feature store latency by 95% - here's how we did it The silent killer in many ML pipelines? Inefficient feature stores. We discovered this the hard way when response times jumped from 50ms to 2 seconds in production. Key optimizations that transformed our performance: Implemented feature computation graphs to eliminate redundant calculations Deployed smart multi-level caching with hot/warm/cold feature strategies Built intelligent batch processing with dynamic sizing The results were dramatic: ✨ Feature computation 📉 CPU utilization 💰 Storage costs 🎯 Model accuracy Check out our detailed technical breakdown and implementation guide here: https://lnkd.in/d3pf6e8P

4

1 Kommentar

Join us as we unpack the future of AI with Carsten Kraus in this follow-up article from Launch's #NavigatingAbroad series. Learn about the impactful role of small language models in making AI more accessible and the intricate dynamics of reinforcement learning. What implications will these technological advances have on the future landscape of AI? #AIUnpacked #FutureOfAI #LanguageModels #ReinforcementLearning #TechInnovation

6

A great read on the performance and financials of LLM fine-tuning. Really hoping the outcome is a tool that helps overcome the biggest elephant in the room, Mr. Cost! #llm #finetuning #finance #computaioncosts

2

1 Kommentar

💬  Learn how to build a custom chatbot with LLMs + RAG and scalable workflows 🤖 RSVP now to join our Chief ML Engineer, Niels Bantilan, for a live workshop online covering: - Using LLMs - Building RAG system for information retrieval - Prompt engineering - Creating scalable workflows using Union.ai (powered by Flyte) Who should attend: Anyone interested in building custom chatbots with LLMs and best MLOps practices should attend. This workshop is designed to be approachable for most skill levels. Familiarity with machine learning and Python is strongly encouraged but not required. By the end of this workshop, you'll be able to build a custom chatbot using LLMs, RAG, and scalable AI workflows for large language models (LLMs). 🔗 https://lnkd.in/gnTajH3Z #llm #llms #rag

13

1 Kommentar

Wondering how various parameters affect Retrieval-Augmented Generation (RAG) performance in practice ⁉️ New blog episode presents hands-on exploration of factors like chunk size, overlap, and context size, examining their impact on the LLM’s ability to answer questions. 🛠️ It also delves into error analysis and the cost breakdown of running the full pipeline, including token usage and pricing considerations.💡 Read now and thank Luka Panić later for nailing it again! 🏆🌟➡️ bit.ly/3A3jSgC #PixionAiBlogSeries #AI #LLM #RAG

7

🧠 What’s a Semantic Cache in RAG? Here’s What We’ve Learned Semantic cache is a time and cost saving method for Retrieval Augmented Generation (RAG). Instead of always running a full vector search, a semantic cache stores embeddings of frequently asked queries and their responses. This speeds up your RAG system by skipping redundant searches for similar queries. How do you implement it? Step 1: Store history of answered questions, their embeddings and the LLM response in a semantic cache table. Step 2: When a new query comes in, compare it against the cache table using vector similarity. Step 3: If there’s a match with score over a certain threshold, serve the cached response. If not, run a fresh retrieval and add it to the cache. Now, let’s talk vector search methods: 1️⃣ Cosine Similarity: Measures angle between vectors, works well for normalized data. 2️⃣ Euclidean Distance: Measures straight-line distance but can be skewed by vector magnitude. 3️⃣ Dot Product: Our favorite—it factors in vector size and performs best for our use cases. Want to see how semantic caching powers a RAG system in real life? Check out our demo at https://lnkd.in/g2jFuHTT. 🚀

2

Is Postgres all you need? Is a vector just a data type? That's a tough question. On one hand, dedicated vector databases have a lot of catching up to do in terms of almost 30 years of functionality. On the other, there are advantages to reimagining the architecture factoring in all we know in 2024. The good news with txtai is that it's capable of working with multiple setups. It can persist data to Postgres. It can store data in Faiss + SQLite. It can also integrate content with other vector databases. The idea is to have everything needed to get started fast and be flexible to change as the requirements and landscape evolves. https://lnkd.in/dxWDeey

13

For all the wizardly power of Large Language Models (LLMs), they cannot reliably handle math. In our world of strategic insights, we need both deep conversations and rigorous statistics, integrated into one cohesive story. Tools like ChatGPT and Claude are game changers when it comes to analyzing the "conversation" part of the equation. To overcome their inherent weaknesses in reliably computing math and statistics, we utilize something they are now very good at, Python programming. By leveraging Python's robust libraries, LLMs can bypass their mathematical constraints and execute precise computations. Whether you're a data scientist, financial analyst, or researcher, you can learn how integrating LLMs with Python can enhance your analytical tasks, from simple calculations to complex forecasting models. Our latest piece shows how allowing LLMs to utilize Python programming democratizes access to advanced computational tools, enabling users with limited programming expertise to find insights from sophisticated mathematical and statistical techniques. #ArtificialIntelligence #MachineLearning #DataScience #PythonProgramming #LLMs #MathInAI #DataAnalysis https://lnkd.in/gq-QWqig

7

New Blogpost: Fine-tuning LLMs for Multi-turn Conversations In this deep dive we cover: 🎛️ How to fine-tune with conversation data. 🎭 Implement loss masking to improve model performance. 🐍 Full code example fine-tuning Llama 3.1 8B on CoQA dataset. Read here: https://lnkd.in/gFxbpia6

20

🚀 New in txtai 7.3: Streaming RAG. This feature builds off streaming LLMs which iteratively return chunks of content as a stream vs waiting for the entire generation call. Streaming is supported with Transformers, llama.cpp and LLM APIs (i.e. GPT-4, Claude, Ollama) Try it out with txtai's new RAG application. Plug in a new txtai embeddings index to use this application with your own data. Docker image: https://lnkd.in/d-fJiVXR RAG Application: https://lnkd.in/e82fddyY

5

🚀 Exciting News! 🚀 We are thrilled to announce that Kumo AI's position paper "Relational Deep Learning: Graph Representation Learning on Relational Databases," has been accepted at #ICML 2024! 🎉 Join us in Vienna from July 21-27, where our very own Matthias Fey will be presenting this exciting new research area. 🔗 https://icml.cc/ In collaboration with Stanford University, we introduce Relational Deep Learning (#RDL), a new research area for applying deep learning on relational databases by viewing them as temporal, heterogeneous graphs. This method eliminates the need for manual feature engineering, leading to more accurate models built faster. 🚀 We’ve also developed #RelBench, a benchmark suite to facilitate research in this new area, using diverse datasets from Stack Exchange discussions to Amazon book reviews. 📚 Stay tuned for more updates on our presentation format, and check out our paper here: https://lnkd.in/dk24vbSg #ICML2024 #MachineLearning #AI #GraphNeuralNetworks #RelationalDeepLearning #DataScience #KumoAI #StanfordCollaboration

38