Aniq Kassam’s Post

An in-depth masterclass to learn and few takeaways; 1. How to make crucial architecture decisions, select optimal model types, configure hyperparameters, and curate quality training data. 2. Discover professional techniques for pre-training, iterative fine-tuning, and rigorous model evaluation. 3. A walkthrough of the complete LLM lifecycle, this masterclass empowers with hands-on skills to build, refine, and deploy large language models with confidence. #buildmodels #AI #learningisfun

[New on our blog] Continual Learning: Methods and Application by Mateusz Wójcik TL;DR → In many machine-learning projects, the model has to frequently be retrained to adapt to changing data or to personalize it. → Continual learning is a set of approaches to train machine learning models incrementally, using data samples only once as they arrive. → Methods for continual learning can be categorized as regularization-based, architectural, and memory-based, each with specific advantages and drawbacks. → Adapting continual learning is an incremental process, from carefully identifying the objective over implementing a simple baseline solution to selecting and tuning the continual learning method. → The key to continual learning success is identifying the objective, choosing the right tools, selecting a suitable model architecture, incrementally improving the hyperparameters, and using all available data. — (link to the full article in the comments) #ML #MLOPs #ContinualLearning

📚Good read📚 Destin Gong of visual-design.net provides an extensive walkthrough in Towards Data Science on how to develop more collaborative, reproducible and reusable ML code. This article is fully equipped with examples, code and equally important — the logic that supports each tip provided. Check it out!

A new blog post detailing my journey with Copilot and other Large Language Models (LLMs) and how they've revolutionized my coding process. It's been an incredible experience seeing firsthand how these tools can enhance productivity, creativity, and efficiency in coding. 🚀💻🔍🛠️✨👨💻🔗 #Coding #Technology #Innovation #LLMs #Copilot #DeveloperCommunity

Hearing the same question over and over again either in person or on the various networks, I decided to share my thoughts and of course, to get other points of view, suggestions and experiences. "How to introduce LLMs and the rapidly evolving ecosystem to engineers in my company with no background on the subject?", they ask. My quick answer (no presentation or template, sorry) is to try to build it as a workshop and cover the following: 1. Lightweight explanation about the mechanics behind how LLMs work (yes, the transform architecture as well - they're engineers, they should get it. At least high level). There's that cool GPT-2 excel sheet which is a great learning tool. Also mention the various LLMs out there, open vs closed source, and how they are currently benchmarked. 2. Prompt engineering, tactics and methods such as few shot learning, chain of thought, etc. Use a lot of live examples at this point to demonstrate how leveraging different methods improve the outcome. Since it's a workshop you can pause for a challenge aiming to achieve something with an LLM that is usually done by combining those methods. 3. Next I would suggest exposing them to embeddings, vector databases and the RAG concept. It has become such a common and cardinal practice, you shouldn't stop before going over this as well (but you might want to split the session into two). So after a short theoretical explanation about embeddings and vector DBs, followed by examples, introduce RAG and immediately start another hands on practice. There are so many resources online that will guide any developer to implement a system that takes documents, chunks them up, creates embeddings, stores them in a database and finally enables you to query the data, all in under one hour. Tip: Use company-specific data for this, such as your knowledge base, your blog posts, your product guides, history of a public Slack channel, etc. 4. Finally, talk about fine tuning LLMs (I wouldn't go too deep into this one), and if you haven't explained before, that creating embeddings is also done using a model, which can be a fine-tuned foundation model's embedding model. To sum it up, talk about the opportunities for your company/products/services that emerge given such capabilities. Sometimes "newcomers" have the best ideas! What would you add or do different? Have you tried something similar, how did it go?

Effective feature engineering is vital for machine learning but is often time-consuming and complex. It relies on domain expertise and can be impacted by changing business strategies. Ad hoc feature engineering causes inefficiency and duplicated efforts. It also risks training–serving skew or data leakage by introducing label information into the model input pipeline. A feature store can solve these issues by providing a centralized repository for feature datasets. This store improves consistency, speeds up development, and enables better feature management, including versioning, documentation, and access control.

🎉 Exciting news! Today I earned my "Create a Composed Document Intelligence Model" badge! 🚀 I'm so proud to be celebrating this achievement and hope it inspires you to start your own @MicrosoftLearn journey! 📚✨ 🔍 Crafting a composed document intelligence model has been an amazing experience: - Enhanced my ability to analyze and manage complex documents! 📄🤖 - Supercharged my productivity with advanced AI solutions. ⏱️⚙️ - Mastered the integration of innovative technology into everyday tasks. 💻🔗 🌟 This badge is a testament to hard work, dedication, and a lot of fun along the way! 🌟 Why should you embark on your @MicrosoftLearn journey? 🤔 - Endless opportunities to enhance your skills and grow. 🌱🚀 - Connect with a dynamic community of fellow learners. 🌐💬 - Gain practical experience with state-of-the-art technology. 🔧🎓 Ready to dive in and earn your own badge? 🏅 What skills are you eager to develop next? Share your thoughts and let's support each other on this journey! 💪👏 #MicrosoftLearn #AI #DocumentIntelligence #ContinuousLearning

What is a Mixture-of-Experts (MoE)? A Mixture of Experts (MoE) is like having a team of specialists, each one skilled at handling different parts of a complex task. Imagine breaking down a big project into smaller tasks and assigning each one to an expert who excels in that specific area. How Does It Work? Think of MoE as a smart version of a popular machine learning model called a transformer. In an MoE, there is a special component called an MoE block. This block includes several experts (which are like mini-networks) and a gating function that decides which expert should handle each part of the data. Key Components: Experts: These can be simple feedforward networks or even large language models (LLMs). Gate/Router: This part uses a softmax function to determine which expert should process each incoming piece of data. Why is MoE Special? Specialization: Different experts can focus on different tasks. For example, one might be great at coding, another at math, and another at writing. Efficiency: Experts can work in parallel on different GPUs, making the process faster. Scalability: Each token (piece of data) is processed by the most relevant expert(s), adding more learnable parameters without increasing the cost of running the model. This is a huge advantage! Recommended Reading: GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding (2020): [Read Here](https://lnkd.in/gcembJrZ) Towards Understanding MoEs (2022): [Read Here](https://lnkd.in/gG6ZFBDt) Mixture-of-Experts Meets Instruction Tuning (2023): [Read Here](https://lnkd.in/gwueAXiM) If you want to run Mistral MoE, dive into these resources and get started! 🚀

Turn distributed data into a competitive edge with #HPE Swarm Learning. 🚀 Want to learn how? Review this comprehensive white paper to find out! You'll learn what swarm learning is, how to apply it, and how it will benefit your business. 🙌 If you have additional questions, MainMicro can help.