InstructLab

🚀 Exciting News! Call for Proposals for DevConf.IN 2025 is Now Open! 🚀 We’re calling on all tech students, enthusiasts, developers, thought leaders, customers, and partners to submit proposals for DevConf.IN 2025, happening at MIT World Peace University, Pune. Here’s what we’re looking for: 🔹 Cloud, Edge, and Platform Technologies 🔹 AI, Data Science, and Emerging Tech 🔹 Security and Sustainable Computing 🔹 Open Source Innovation and Leadership 🔹 User Experience and Design Engineering 💡 Share your expertise with a global audience at one of the biggest open-source events of the next year! 🗓 Important Dates: • CfP Submission Deadline: December 15, 2024 • Notifications: January 13-17, 2025 • Event: February 28 - March 1, 2025 🔗 Submit here: https://lnkd.in/dvri9JGm #OpenSource #DevConfIN #TechInnovation #CallForProposals #CloudComputing #AI #DataScience #CyberSecurity #DesignEngineering #RedHat #OpenSourceLeadership #CommunityDriven #TechCommunity #TechConference #Innovation #DevConf2025

🎃 For Halloween this year, I’m dressed up as InstructLab, an innovative open-source #AI project developed by IBM and Red Hat! If you’re wondering “what is InstructLab?”, don’t worry - I’ve got you covered. In simple terms, InstructLab solves major challenges faced by businesses in fine-tuning and deploying large language models (LLMs) for their #genAI applications. 𝘾𝙤𝙨𝙩𝙪𝙢𝙚 𝙚𝙭𝙥𝙡𝙖𝙞𝙣𝙚𝙙: 🌲 “Taxonomy”: Hanging around my neck is a depiction of InstructLab’s organizational tree structure. Using this, businesses can easily and intuitively add the new knowledge and skills they want added to an existing AI model. 🍎 “Teacher model”: AI is built on data, but collecting a giant dataset is time-consuming, difficult and costly. In InstructLab, you can add in just a handful of examples (less than 10!)... and the teacher model will go out and synthetically create hundreds or thousands more! In my costume, the “teacher” is in front of a backdrop of model, which is a giant matrix of numbers. 🤔 “Critic model”: Wait a sec… How do we ensure the synthetic examples are any good? A critic model then evaluates this data to filter out those that don’t meet your predefined criteria (i.e., irrelevant or potentially harmful). Basically, this ensures high-quality and contextual grounding. In my costume, the “critic” is again in front of a backdrop of a “model” (matrix of numbers) and “criticism” (thumbs up/down). 𝙒𝙝𝙖𝙩 𝙘𝙤𝙢𝙢𝙤𝙣 𝙥𝙧𝙤𝙗𝙡𝙚𝙢𝙨 𝙙𝙤𝙚𝙨 𝙄𝙣𝙨𝙩𝙧𝙪𝙘𝙩𝙇𝙖𝙗 𝙨𝙤𝙡𝙫𝙚?: 💰 Vs pre-training: Using InstructLab over an existing model is waaaaaay less expensive than purchasing a GPU farm and creating your own model. 🎵 Vs fine-tuning: InstructLab solves the major problem of “catastrophic forgetting”, or learning new things without forgetting previous lessons. 💬 Vs prompt tuning or RAG: InstructLab actually trains the model whereas prompt tuning and Retrieval Augmented Generation (RAG) do not. I also need to add a BIG thank you to the amazing Bonnie Balu for being my executive curator, structural engineer, seamstress, chief voice of reason, and probably about 37 other roles!

Excited to share my latest tutorial on InstructLab UI! This hands-on guide demonstrates how anyone—from AI Engineers and developers to domain experts in fields like medicine or law—can easily fine-tune open-source Large Language Models (LLMs) like IBM Granite. Have you ever interacted with an LLM on a topic you know well but felt unsatisfied with its responses? InstructLab empowers you to contribute your domain expertise back to the model, whether through open-source contributions for public use or Inner Source within your enterprise. No AI or ML experience, nor extensive hardware resources, are required. You can make contributions directly from your laptop, helping to democratize AI. The recently launched InstructLab UI offers an intuitive interface that simplifies contributions by eliminating the complexity of YAML structures, validation rules, and GitHub pull requests. Whether you're contributing knowledge or skills that teach the model how to perform specific tasks, InstructLab UI allows you to focus on enhancing the LLM's capabilities without feeling overwhelmed by tools and processes. I want to express my sincere gratitude to Red Hat and IBM colleagues (Anil Vishnoi, Susan Malaika, and Suhas Kashyap) for their guidance and expertise in reviewing and contributing to this tutorial. A special thank you to IBM Developer's outstanding technical editor Michelle Corbin for her great efforts in editing the tutorial. #AI #InstructLab