Sarfarazuddin Syed’s Post

Self-attention as a directed graph! Self-attention is at the heart of transformers, the architecture that led to the LLM revolution that we see today. In this post, I'll clearly explain self-attention & how it can be thought of as a directed graph. Keep Swiping through the images below!👇 For those interested in Ai Engineering, I would also encourage you to check Lightning AI's ⚡️ LLM learning Lab. A curated collection of blogs, tutorials, and how-to videos on: - Training - Fine-tuning - And deploying LLMs 🚀 Check this out👇  https://lnkd.in/d-X-gcCe _____________ That's a wrap, hope you enjoyed reading! If you're interested in: • LLMs 🧠 • MLOps 🛠 • Python 🐍 • And Machine Learning ⚙️ Find me → https://lnkd.in/em_V4unu ✔️ Everyday, I share tutorials on above topics! _____________ Cheers!! 🙂 #artificialintelligence #llms #machinelearning

A smart and simple way to explain the attention mechanism .

Level Up Your ML Game with These 6 Must-Subscribe YouTube Channels! 🚀 1️⃣ Sentdex by Harrison Kinsley – Hands-on Python & ML tutorials. 2️⃣ Data School – Simplifying data science for everyone. 3️⃣ DeepLearning.AI – Learn from industry pioneers. 4️⃣ Machine Learning with Phil Tabor – Practical ML insights and tips. 5️⃣ Jeremy Howard – Advanced deep learning and fast ai. 6️⃣ Two-Minute Papers – Breaking down AI research in minutes. ✨ Which one’s your favorite? Let me know! Save this post to revisit later! 🔖 ====================== Experts predict 2025 will be the defining year for Gen AI implementation. Want to be ahead of the curve? Join Build Fast with AI’s Gen AI Launch Pad 2025 - your accelerated path to mastering AI tools and building revolutionary applications. Register now: https://lnkd.in/dgCEJ9Ed

Day 37 of 1000 Days of AI/ML Today, I explored advanced optimization techniques and practical tools for ML! 🤖 From CS50's AI Course - Week 3 Optimization (https://lnkd.in/gqp-yMVV) 1. Backtracking Search: A recursive algorithm assigning values to variables while satisfying constraints. It backtracks on conflicts. 2. Inference with AC-3: Combining backtracking with AC-3 enforces arc-consistency during the search, enhancing efficiency. 3. Heuristics in Backtracking: MRV (Minimum Remaining Values): Prioritize variables with the smallest domain to minimize backtracking. Degree Heuristic: Select variables with the most constraints for efficient problem-solving. LCV (Least Constraining Values): Pick values that limit other variables the least, keeping flexibility high. 📚 From Machine Learning YouTube Course (https://lnkd.in/gNhzCWyC) 1. Anaconda 🐍 Installed Anaconda to simplify Python package management and virtual environment setup. 2. Kaggle: Ran notebooks and uploaded datasets 3. Google Colab: Knowing it offers free TPU/GPU for training models and how to integrate large datasets from Kaggle in colab. Reflection & Plans: Today, I learned about optimization with backtracking and heuristics while setting up Anaconda, Kaggle, and Colab for an efficient ML workflow . Tomorrow, I’ll start looking into Week 3 problem set Crossword and machine learning playlist from youtube. #AIJourney #1000DaysOfAI #Optimization #Heuristics #Kaggle #GoogleColab #Anaconda

💻 Exploring the Foundations of Predictive Machine Learning 💡 We recently dove into some of the key elements that go into developing a predictive machine learning model. At Sauce, we handle massive datasets regularly, and our priority is always adding as much value to our clients’ projects as possible. 🌟   In our latest session, we focused on the critical decisions and processes that happen before applying a model to data. Here’s what we covered: 🔎 ML Types: Supervised, Unsupervised, and Reinforcement Learning – with a deep dive into predictive modelling using supervised learning (regression & classification). 🔑 Key Concepts: Loss functions, hyper-parameters, feature engineering, discretisation, train-test splits, and scaling/normalisation. ⚖ Choosing the Right Model: We discussed starting with simple, interpretable models and gradually working towards more complexity. We also explored getting started with both Python’s Jupyter Notebooks and Elixir Livebook. While Python is the industry standard for ML, we’re excited about the rapid evolution of Elixir’s Nx and Axon libraries. Given Elixir’s design for concurrency, there’s great potential for it to become an efficient ML platform! #MachineLearning #DataScience #Python #Elixir #AI #TechInnovation

𝑫𝒊𝒗𝒊𝒏𝒈 𝒊𝒏𝒕𝒐 𝑷𝒚𝒕𝒉𝒐𝒏 𝑴𝒂𝒄𝒉𝒊𝒏𝒆 𝑳𝒆𝒂𝒓𝒏𝒊𝒏𝒈 ! I recently explored "Python Machine Learning: A Beginner's Guide to Scikit-Learn," and it’s packed with gems for beginners and enthusiasts alike. Here’s a quick roundup of my takeaways: 1. Python’s simplicity, paired with powerful libraries such as TensorFlow, and Pandas, makes it a top choice for machine learning. The language’s readability and large community support are key for both development and troubleshooting. 2. Data preparation is essential before building any model. Data cleaning, exploration, and feature engineering ensure accuracy, and Scikit-learn makes this easier with tools for data splitting and feature transformation. 3. Supervised and unsupervised learning are foundational. Supervised models use labeled data, while unsupervised ones identify patterns in raw data. This distinction is crucial when choosing the right approach for each task. 4. Ensemble learning enhances predictions by combining models. Techniques like bagging, boosting, and stacking each have strengths and knowing when to use them can make all the difference in results. 5. Real-world applications span from NLP and computer vision to fraud detection and robotics. Seeing these examples illustrates the vast impact of ML and sparks ideas for applying these concepts to future projects. For anyone starting in machine learning, this book is a fantastic resource! 💡 Follow Alisha Surabhi for more! #MachineLearning #Python #DataScience #LearningJourney #AI

𝐓𝐞𝐧 𝐏𝐲𝐭𝐡𝐨𝐧 𝐋𝐢𝐛𝐫𝐚𝐫𝐢𝐞𝐬 𝐭𝐨 𝐁𝐨𝐨𝐬𝐭 𝐌𝐚𝐜𝐡𝐢𝐧𝐞 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 Machine Learning workflows often rely on popular libraries like Scikit-learn, TensorFlow, and PyTorch. But there’s a treasure trove of lesser-known Python libraries that 𝐜𝐚𝐧 𝐞𝐥𝐞𝐯𝐚𝐭𝐞 𝐲𝐨𝐮𝐫 𝐩𝐫𝐨𝐣𝐞𝐜𝐭𝐬 𝐭𝐨 𝐭𝐡𝐞 𝐧𝐞𝐱𝐭 𝐥𝐞𝐯𝐞𝐥! Here are 10 hidden gems I recently explored: 🔍  𝐈𝐦𝐛𝐚𝐥𝐚𝐧𝐜𝐞𝐝-𝐥𝐞𝐚𝐫𝐧: Tackle class imbalances in fraud detection or medical diagnostics. ⚙️ 𝐓𝐏𝐎𝐓: Let genetic algorithms automate your model pipelines. 📊 𝐅𝐞𝐚𝐭𝐮𝐫𝐞-𝐞𝐧𝐠𝐢𝐧𝐞: Simplify preprocessing with intuitive tools for encoding and imputing. 📈  𝐘𝐞𝐥𝐥𝐨𝐰𝐛𝐫𝐢𝐜𝐤:Visualize and understand your models better. ⚡ 𝐇𝐮𝐦𝐦𝐢𝐧𝐠𝐛𝐢𝐫𝐝: Convert models for blazing-fast GPU inference. 🎯 𝐎𝐩𝐭𝐮𝐧𝐚:  Tune hyperparameters intelligently for resource-sensitive applications. 🌊 𝐑𝐢𝐯𝐞𝐫: Streamline online learning for real-time predictions. ⏱️ 𝐏𝐲𝐂𝐚𝐫𝐞𝐭:: Build and deploy models faster with minimal coding. 🔗  𝐏𝐨𝐦𝐞𝐠𝐫𝐚𝐧𝐚𝐭𝐞: Dive into probabilistic reasoning with Bayesian Networks. 💡  𝐌𝐥𝐱𝐭𝐞𝐧𝐝:: Experiment with advanced techniques like stacking and association rules. These libraries go beyond traditional tools, addressing niche challenges in 𝐟𝐢𝐧𝐚𝐧𝐜𝐞, 𝐡𝐞𝐚𝐥𝐭𝐡𝐜𝐚𝐫𝐞, 𝐈𝐨𝐓, 𝐚𝐧𝐝 𝐛𝐞𝐲𝐨𝐧𝐝. Whether you're optimizing for speed, explainability, or automation, there’s something here for everyone! 👉 Have you used any of these libraries? Which one transformed your ML workflow? Share your thoughts below! Let’s connect, collaborate, and innovate. Together, we can shape the future of tech! 🌐 Follow Sanjay Nandakumar for such posts! #MachineLearning #Python #AI #DataScience #DeepLearning #Statistics #Artificialntelligence

Can you guess which are the most popular ML libraries in production today? Tensorflow is now a tale of the past with only 8%! Sklearn reaches the top with 35% closely followed by Pytorch with 32% - other contenders are XGBoost with 7% and Catboost with 6% 💻 We are uncovering great insights as part of our survey on The State of Production ML in 2024; please contribute to this valuable investigation on machine learning tools and platforms used in your production ML development. Your input will help create a comprehensive overview of common practices, tooling preferences, and challenges faced when deploying models to production, ultimately benefiting the entire ML community 🚀 Survey: https://lnkd.in/eFhTYvqm Results: https://lnkd.in/eHk98iC2 -- If you liked this article you can join 60,000+ practitioners for weekly tutorials, resources, OSS frameworks, and MLOps events across the machine learning ecosystem: https://lnkd.in/eRBQzVcA #ML #MachineLearning #ArtificialIntelligence #AI #MLOps #AIOps #DataOps #augmentedintelligence #deeplearning #privacy #kubernetes #datascience #python #bigdata

Can you guess which are the most popular ML libraries in production today? Tensorflow is now a tale of the past with only 8%! Sklearn reaches the top with 35% closely followed by Pytorch with 32% - other contenders are XGBoost with 7% and Catboost with 6% 💻 We are uncovering great insights as part of our survey on The State of Production ML in 2024; please contribute to this valuable investigation on machine learning tools and platforms used in your production ML development. Your input will help create a comprehensive overview of common practices, tooling preferences, and challenges faced when deploying models to production, ultimately benefiting the entire ML community 🚀 Survey: https://lnkd.in/eb4udcDs Results: https://lnkd.in/eRWa_AYC -- If you liked this article you can join 60,000+ practitioners for weekly tutorials, resources, OSS frameworks, and MLOps events across the machine learning ecosystem: https://lnkd.in/eRBQzVcA #ML #MachineLearning #ArtificialIntelligence #AI #MLOps #AIOps #DataOps #augmentedintelligence #deeplearning #privacy #kubernetes #datascience #python #bigdata

How LLMs understand relative positions of input words, clearly explained! RoPE (Rotary Positional Embeddings) revolutionised the way positional information is encoded in LLMs and it's widely used by models like Llama-3. Today, I'll clearly explain what they are & how positional embeddings evolved over time. Here's what we cover: - What are positional embeddings - Absolute & relative positional embedding - Limitations of Absolute & relative positional embedding - Rotary Positional Embeddings & what makes them so powerful The very ability of RoPE to capture relative positions while being parameter efficient has been key in development of very long-context LLMs. So, let's dive in and visually understand how it works! __________________ If you interested in: - Python 🐍 - ML/MLOps 🤖 - AI Engineering ⚙️ 1. Find me → https://lnkd.in/em_V4unu ✔️ 1. Sign up for our newsletter for in-depth lessons and get a FREE eBook with 150+ core DS/ML lessons: https://lnkd.in/gB7yHExC