We introduce Data Orchestration Hub, a management service that makes it easy to build an analytics or machine learning platform on data sources across regions to unify data lakes. Easy to use wizards connect compute engines, such as Presto or Spark, to data sources across data centers or from a public cloud to a private data center. In this session, you will witness the use of “The Hub” to connect a compute cluster in the cloud with data sources on-premises using Alluxio. This new service allows you to build a hybrid cloud on your own, without the expertise needed to manage or configure Alluxio.
We introduce Data Orchestration Hub, a management service that makes it easy to build an analytics or machine learning platform on data sources across regions to unify data lakes. Easy to use wizards connect compute engines, such as Presto or Spark, to data sources across data centers or from a public cloud to a private data center. In this session, you will witness the use of “The Hub” to connect a compute cluster in the cloud with data sources on-premises using Alluxio. This new service allows you to build a hybrid cloud on your own, without the expertise needed to manage or configure Alluxio.
Video:
Presentation Slides:
Complete the form below to access the full overview:
.png)
Videos
TorchTitan is a proof-of-concept for Large-scale LLM training using native PyTorch. It is a repo that showcases PyTorch's latest distributed training features in a clean, minimal codebase.
In this talk, Tianyu will share TorchTitan’s design and optimizations for the Llama 3.1 family of LLMs, spanning 8 billion to 405 billion parameters, and showcase its performance, composability, and scalability.
As large-scale machine learning becomes increasingly GPU-centric, modern high-performance hardware like NVMe storage and RDMA networks (InfiniBand or specialized NICs) are becoming more widespread. To fully leverage these resources, it’s crucial to build a balanced architecture that avoids GPU underutilization. In this talk, we will explore various strategies to address this challenge by effectively utilizing these advanced hardware components. Specifically, we will present experimental results from building a Kubernetes-native distributed caching layer, utilizing NVMe storage and high-speed RDMA networks to optimize data access for PyTorch training.