You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@@ -1463,12 +1463,12 @@ <h5 align="left">Peter Grabowski, Lead of Gemini Applied Research, Google</h5>
1463
1463
<divclass="col-md-12 v-center">
1464
1464
<h6>Talk Abstract</h6>
1465
1465
<p>
1466
-
Want to get started with LLMs? This lecture will cover an introduction to language modeling and prompt engineering, example use cases and applications, and a discussion of common considerations for LLM usage (cost, efficiency, accuracy, bias).
1466
+
Coming soon!
1467
1467
</p>
1468
1468
1469
1469
<h6>Speaker Bio</h6>
1470
1470
<p>
1471
-
Peter leads the Gemini Applied Research group, focused on developing fast, efficient, and scalable models in partnership with DeepMind, Search, Ads, Cloud, and other teams across Google. Prior to that, he led a group focused on Google's Enterprise AI, worked on making the Google Assistant better for Kids, and led the data integration / machine learning team at Nest. Peter loves to teach, and is a member of the faculty at UC Berkeley's School of Information, where he teaches courses focused on Deep Learning and Natural Language Processing.
1471
+
Christopher Bishop is a Microsoft Technical Fellow and a member of Microsoft Research AI for Science. Chris obtained a BA in Physics from Oxford, and a PhD in Theoretical Physics from the University of Edinburgh, with a thesis on quantum field theory. After his PhD he joined the Theoretical Physics Division of Culham Laboratory where he conducted research into the physics of magnetically confined fusion plasmas. During this time, he developed an interest in machine learning and became Head of the Applied Neurocomputing Centre at AEA Technology. He was subsequently elected to a Chair in the Department of Computer Science and Applied Mathematics at Aston University, where he set up and led the Neural Computing Research Group. He joined Microsoft in 1997 and was Lab Director of Microsoft Research Cambridge from 2015 until 2022 when he founded the new AI for Science team. At Microsoft Research, Chris oversees a global portfolio of research, focussed on machine learning for the natural sciences.
<h4align="left">Introduction to LLM Post-Training</h4>
1503
-
<h5align="left">Maxime Labonne, Head of Post-Training, Liquid AI</h5>
1502
+
<h4align="left">Massively Parallel Training</h4>
1503
+
<h5align="left">Mathias Lechner, Co-Founder and Chief Technology Officer, Liquid AI</h5>
1504
1504
</div>
1505
1505
</div>
1506
1506
</div>
@@ -1509,12 +1509,12 @@ <h5 align="left">Maxime Labonne, Head of Post-Training, Liquid AI</h5>
1509
1509
<divclass="col-md-12 v-center">
1510
1510
<h6>Talk Abstract</h6>
1511
1511
<p>
1512
-
In this talk, we will cover the fundamentals of modern LLM post-training at various scales with concrete examples. High-quality data generation is at the core of this process, focusing on the accuracy, diversity, and complexity of the training samples. We will explore key training techniques, including supervised fine-tuning, preference alignment, and model merging. The lecture will delve into evaluation frameworks with their pros and cons for measuring model performance. We will conclude with an overview of emerging trends in post-training methodologies and their implications for the future of LLM development.
1512
+
This lecture talks about how to scale training of deep neural networks to thousands of GPUs. It begins by motivating why GPUs are essential for training (comparing FLOPs of GPUs vs CPUs) and why scaling to larger models and datasets improves performance, drawing on scaling laws from LLaMA and Kaplan et al. The talk then explores the memory requirements of training and techniques to reduce them, including activation checkpointing and offloading. The bulk of the lecture covers parallelism strategies: data parallelism, tensor parallelism, pipeline parallelism, and sequence/context parallelism, as well as sharding approaches like DeepSpeed ZeRO and FSDP. It also touches on sparsity through Mixture of Experts and expert parallelism. Throughout, network bandwidth is highlighted as a key bottleneck. The lecture concludes with a case study of LFM2 showing how these techniques combine in practice.
1513
1513
</p>
1514
1514
1515
1515
<h6>Speaker Bio</h6>
1516
1516
<p>
1517
-
Maxime Labonne is Head of Post-Training at Liquid AI. He holds a Ph.D. in Machine Learning from the Polytechnic Institute of Paris and is a Google Developer Expert in AI/ML. He has made significant contributions to the open-source community, including the LLM Course, tutorials on fine-tuning, tools such as LLM AutoEval, and several state-of-the-art models like NeuralDaredevil. He is the author of the best-selling books “LLM Engineer’s Handbook” and “Hands-On Graph Neural Networks Using Python”.
1517
+
Mathias Lechner is Co-Founder and Chief Technology Officer (CTO) at Liquid AI, as well as a Research Affiliate at the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT, where he collaborates with Prof. Daniela Rus. He completed his PhD in 2022 at the Institute of Science and Technology Austria (ISTA), under the supervision of Tom Henzinger. Before his PhD, he earned his master’s (2017) and bachelor’s (2016) degrees in Computer Science from the Vienna University of Technology (TU Wien).
The potential of AI in biology is immense, yet its success is contingent on interfacing effectively with wet-lab experimentation and remaining grounded in the system, structure, and physics of biology. I will share how, at Microsoft Research, we are developing new AI systems that help us better understand and design biology via generative design and interactive discovery. I will focus on Generative AI models for the design of novel and useful biomolecules, expanding our ability to engineer new proteins for therapeutic, biological, and industrial applications and beyond.
1558
+
Coming soon!
1559
1559
</p>
1560
1560
1561
1561
<h6>Speaker Bio</h6>
1562
1562
<p>
1563
-
Ava Amini is a Senior Researcher at Microsoft, where she develops new AI technologies for precision biology and medicine. She completed her PhD in Biophysics at Harvard University and her BS in Computer Science and Molecular Biology at MIT and has been recognized by the National Academy of Engineering, the National Science Foundation, TEDx, Venture Beats, and the Association of MIT Alumnae, among others, for her research. Ava is passionate about AI education and outreach -- she is a lead organizer and instructor for MIT Introduction to Deep Learning, where she has taught AI to 1000s of students in-person and over 100,000 globally registered students online, garnering more than 11 million online lecture views, and served as a co-founder and director of MomentumAI, which taught all-expenses-paid education programs for high schoolers to learn AI.
0 commit comments