🚨 The Problem

Distributed training is full of invisible traps.

Get it wrong, and months of team velocity is gone and you wasted tens or hundreds of thousands of dollars in unoptimized compute.

Most teams guess their way through ZeRO configs, DDP setups, and pipeline logic. That leads to stalled releases and blown budgets.

✅ The Solution

This course is the only live training that teaches modern distributed AI workflows directly from people who have scaled training across tens of thousands of GPUs, and built the foundational knowledge the AI ecosystem is built upon today.

You’ll walk away knowing what to use, when to use it, and why it works, across real production-scale scenarios.

🔥 Built for High-Impact Teams

Engineers: Train large models across 8 to 1,000+ GPUs with precision

Tech leads and CTOs: Make confident decisions on systems, tools, and costs

Founders and startups: Build smarter AI infrastructure that scales without waste

🧩 What Makes This Course Different

* Learn directly from a curated list of 14 world-class experts in the field from Meta, Ray, Snowflake, Hugging Face, and more

* Get five focused workshops with matching hands-on labs for real skills

* Gain access to a private alumni network for continued learning and hiring

* Use $500+ in real compute credits to apply your skills immediately

📝 Guest Experts and Case Studies

Over 14 guest speakers from the top AI teams. Each session ties directly to a problem that will need to be solved when crafting the ideal training scenario.

Applied Track:

Robert Nishihara (Ray)

How to orchestrate GPU training across thousands of nodes effectively

Sami Jaghouar (Prime Intellect)

Building decentralized training systems at a global scale

Tunji Ruwase (Snowflake)

Training long-context models efficiently without exploding memory

Prince Canuma

Running LLMs directly on Apple Silicon for local-first development

Pretraining Track:

Phuc Nguyen (Hugging Face)

Mastering FP8 precision training

Elie Bakouch (Hugging Face)

Advanced MoE and parallelism strategies

Daniel Han (UnslothAI)

How Triton kernels can be an easy optimization win, and other modern practices

Distributed Technique Track:

Sylvain Gugger (Jane Street)

Overview of the ZeRO algorithm

Wanchao Liang (Thinking Machines)

How DTensors helps bring new engineers into understanding distributed training faster

Ferdinand Mom (Hugging Face)

How you should stack parallelism strategies to maximize your training capacity

Less Wright (Meta)

How Async TensorParallelism is necessary to train across clusters of thousands of GPUs

Matej Sirovatka (Hugging Face)

Why Expert Parallelism is a necessity when training MoE models at scale

Marc Sun (Hugging Face)

Why we need new strategies for deploying large models at scale, and how to get there