Deep Dive on LLM Fine-tune In his session, Thomas focuses on understanding the ins and outs of fine-tuning LLMs. We all have a lot of questions during the fine-tuning process. How do you prepare your data? How much data do you need? Do you need to use a high-level API, or can you do this in PyTorch? During this talk, we will try to answer these questions. Thomas will share some tips and tricks on his journey in the LLM fine-tuning landscape. What worked and what did not, and hopefully, you will learn from his experience and the mistakes he made. A Recipe for Training Large Language Models AI models have become orders of magnitude larger in the last few years. Training such large models presents new challenges, and has been mainly practiced in large companies. In this talk, we tackle best practices for training large models, from early prototype to production. What The Kaggle 'LLM Science Exam' Competition Can Teach Us About LLMs This competition challenged participants to submit a model capable of answering science-related multiple-choice questions. In doing so it provided a fruitful environment for exploring most of the key techniques and approaches being applied today by anyone building with LLMs. In this talk, we look at some key lessons that this competition can teach us. Do you really know what your model has learned? Leap Labs demonstrates how data-independent model evaluations represent a paradigm shift in the model development process. All through our dashboard’s beautiful Weights & Biases Weave integration.

Did you know that 87% of AI models never move from lab to production? In fact, this is one of the biggest challenges faced by machine learning teams today. Just because a model excels in a test environment doesn't ensure its success in the real world. Furthermore, as you deploy AI models to production, they often degrade over time, especially with incoming new data. So, how do you know what is causing your AI models to fail? And how do you fix these issues to improve model performance? Let's delve deep into these issues and provide insights on how you and your ML teams can systematically identify and fix model and data issues. Empowered by intelligent workflows, our end-to-end AI platform is built by ML engineers to enhance your model's performance across the entire AI lifecycle, all while unleashing the full potential of robust and trustworthy AI at scale.

Robust Resampling for Time Series Analysis: Introducing tsbootstrap Time series data is ubiquitous across science, engineering, and business. Proper statistical analysis of time series requires quantifying uncertainty in models and forecasts. Bootstrapping is a powerful resampling technique for this, but classical bootstrap methods fail for time series because they disrupt dependencies. This talk introduces tsbootstrap, an open-source Python library implementing state-of-the-art bootstrapping tailored for time series. It covers essential techniques like block bootstrapping along with cutting-edge methods like Markov and sieve bootstrapping. We'll provide an intuitive overview of key time series bootstrapping approaches and demonstrate tsbootstrap functionality for resampling financial data. Users can leverage tsbootstrap for robust confidence intervals, model validation, and distribution-free inference. LLM Security is Real Topic - Safeguard Large Language Model Fortresses!!! Let's highlight a comprehensive agenda addressing the security and ethical challenges of deploying Large Language Models (LLMs). We'll cover topics ranging from LLM security and the risks of prompt injection, denial of service, and model theft to concerns surrounding insecure plugins and overreliance on LLMs. The agenda also includes strategies for tackling these issues and explores data poison mitigation through differential privacy. GeniA: Your Engineering Gen AI Team Member In this talk, Uri introduces a new OSS we've developed named GeniA, and discusses its role as an enabler for the future of platform engineering using Gen AI.