AIFoundry.org

FOSDEM'25 AI DevRoom: Topics Explained! Great dive with Jarek Potiuk into a few topics we highly welcome in the Low-Level AI Engineering and Hacking DevRoom - from models inference engines and quantization to hardware acceleration, llama.cpp governance and new AI open source era principles like freedom to use sowatware as you like. So many interesting perpsectives - we tried to narrow them down for those who are still considering an application. Clocks are ticking! Call for Proposals closes on Dec 1st - we want you in our DevRoom and will be happy to support you in application process or even sponsor your travel to Belgium on Feb 1-2, if necessary. Watch our podcast and learn more about the DevRoom: https://lnkd.in/dVkzwpvi

FOSDEM'25 AI DevRoom: CfP Topics Explained In this podcast, Jarek Potiuk and I will dive into the topics we welcome in speakers' applications - pick your favorite and apply until Dec 1st! - open source AI inference engines - advances in specialized hardware acceleration - model quantization techniques - HPC and computational science for inference, continuous fine-tuning and pre-training - use of free and open source software in the low-level AI engineering community - governance of upstream-downstream relationships (llama.cpp and its downstream, balkanization of llama.cpp/ggml landscape)

"Last year, I attended one talk at FOSDEM that completely changed my trajectory for this year." These are sentiments we’ve heard from many of our participants. FOSDEM isn’t just another tech conference. It’s complete chaos (in the best possible way), where groundbreaking AI engineering and innovation thrive. What makes FOSDEM stand out? Imagine this: Walking into a room buzzing with the most passionate minds in open-source and AI engineering - people whose GitHub profiles you’ve admired - and now you get to meet them in person. That’s what makes FOSDEM special for so many It’s a place where hands-on work takes the spotlight, whether you’re: → Combining models → Discussing distillation techniques → Showing your latest technical ideas Does this sound like your crowd? You’ll feel right at home if you’re: → Working directly with models → Creating practical, real-world AI solutions → Interested in how AI works behind the scenes Don’t just attend - PRESENT your work and meet the community! Best part? It’s all FREE. No registration is required. Just show up and dive in to deepen your learning and skills. So why wait? Join us at FOSDEM to shape the AI future. INTERESTED? DM “FOSDEM” in the comments below!

What’s great about FOSDEM is that all the truly passionate open-source community is there! Whether you're into governance, policy, hardware, low-level hacking, or even niche firmware solutions - there’s a space for everyone. You don't need to register—you just show up It’s free - and still grows and exists for over 20 years. I’ve organized many conferences, but FOSDEM is unique. It’s like what Burning Man used to be—no tickets, just show up and enjoy. It’s not uncommon to spot people in costumes, like someone dressed as an elephant last year. You get to meet incredible people - even while waiting in line for food. The connections made in those moments are unforgettable - even despite the cold weather. Useful swag like umbrellas, hats, and, of course, beer are at your disposal thanks to organizers! If you want to be a part of this - join us in the "Low-Level AI Engineering&Hacking DevRoom!" https://lnkd.in/d3pgefri

The third 'Llama Shaving' podcast edition is here! In it, we tap into how data used to work and how it could be transformed to be effectively utilized by large models. The evolution of data simplified is that, at first, data a collected&preprocessed for humans, then - for static predictable algorhtms, and now it needs to come to a concept digestable by large neural networks. And this is the meta-challenge: we do not fully understand how this can be done. Watch Tanya Dadasheva and Roman Shaposhnik discussing this: https://lnkd.in/enSWDM-d Read the podcast write-up: https://lnkd.in/eHPr235H

Choosing ML Deployment Tools: A Practical Guide Common myths: ? Must be expensive ? Only for big tech ? Needs expert teams Reality check: ? Options for every budget ? Solutions for all levels ? Scalable for any size Quick decision framework: 1?? Small project? FastAPI/Flask 2?? Enterprise? KubeFlow/MLflow 3?? Cloud-first? AWS SageMaker/Azure ML 4?? Need monitoring? MLflow or W&B 5?? Budget tight? Start with open-source Pro Tip: Pick tools that match YOUR needs, not the hype. Start simple, scale later. What deployment tools work for your team? Share below!

Enjoy our 3rd episode of 'Opening AI' podcast - featuring Anatoly, Ainekko CPO, and his product look into the recent AI state report by Sequoia Capital.

We are starting our new podcast series - meet 'Opening AI'! Literally, we try to open up how AI works in these conversations, and the first episode featuring Avi Deitcher will walk you through the process of building systems for production, discussing everything from model deployment to inference, and the challenges of bringing LLMs into the real world. Avi breaks down the concept into three layers, each with its distinct challenges and considerations. Let's explore these layers in detail. Watch https://lnkd.in/dQbK7kgs Read https://lnkd.in/dqEpaQkt

Why community matters.

What most people think about deploying AI in their companies: ? Difficult ? Expensive ? Requires a lot of expertise What it can actually be: ? Easy ? Affordable ? Beginner-friendly Here’s a simple guide for deploying AI models effectively in your business: 1?? Start with a Clear Business Problem Is it customer service? Supply chain optimization? Knowing the problem helps shape your model. 2?? Collect and Clean Your Data AI runs on data – make sure yours is clean, accurate, and structured. 3?? Select the Right AI Model Beginners can leverage pre-built models for quick wins. Explore platforms like Google AutoML, and Hugging Face, or use frameworks like TensorFlow and PyTorch. 4?? Train and Test Your Model Split your data into training and testing sets. Train, test, and tweak to improve accuracy. 5?? Deploy with Cloud Platforms Use beginner-friendly platforms like AWS, Azure, Google Cloud or AI Foundry to deploy and scale your model effortlessly. 6?? Monitor, Improve, Repeat AI is not set and forget. Keep an eye on performance and update your model regularly to stay relevant. Pro Tip: Start small, focus on one problem area, and scale as you gain confidence. AI deployment is a learning process – the more you iterate, the better you get.