LLM Scaling Week

powered by the Yandex School of Data Analysis

  • Yandex School of Data Analysis experts will explain the engineering and mathematical foundations of large language models, how they are trained on GPU clusters, and how to make them faster.

  • The program has finished, but all materials remain available on YouTube and on VK

*The program is held in Russian.
This is not an educational program.
By registering, you agree to the Terms and Conditions

Scaling AI: From experiments to production systems

  • Who's the intensive for?

    • The course is designed for technical university students, researchers, developers, ML engineers, and anyone seeking to scale NLP systems beyond single‑GPU constraints
    •  

    • Practical assignments require familiarity with modern NLP, ML and algorithms such as transformers, GPT‑like models, deep learning, and big data infrastructure
  • What will the participants learn?

    • Experts from the Yandex School of Data Analysis will explain the core concepts and engineering solutions behind modern LLMs
    •  

    • Speakers will demonstrate how to identify inefficiencies and speed up inference
    •  

    • You’ll learn about distributed training and scaling methods used by leading engineers

     

     

What’s in the program

10.11

18:00

Topic 1. Deep Learning Arithmetic

Mikhail Khrushchev

Head of the pretraining group at YandexGPT

Watch on YouTube
Watch on VK
11.11

18:00

Topic 2. Mixture of Experts

Alexander Mazitov

Head of architecture research at YandexGPT

Watch on YouTube
Watch on VK
12.11

18:00

Topic 3.1. Speeding Up Training with FP8 and Triton

Vladislav Savinov

Head of the YandexGPT training 
infrastructure team

Watch on YouTube
Watch on VK

19:20

Topic 3.2. Speeding Up Training with FP8 and Triton

Vladislav Savinov

Head of the YandexGPT training 
infrastructure team

Watch on YouTube
Watch on VK
13.11

18:00

Topic 4. Communications in Distributed Learning and Inference

Stepan Kargaltsev

Lead developer, Yandex

Watch on YouTube
Watch on VK
14.11

18:00

Topic 5.1 Inference Challenges

Roman Gorb

Head of the YandexGPT inference acceleration team

Watch on YouTube
Watch on VK

19:45

Topic 5.2. Inference Challenges. Practice

Roman Gorb

Head of the YandexGPT inference acceleration team

Watch on YouTube
Watch on VK

FAQ

Still have questions?