Increasingly powerful compute accelerators and large training dataset have made the storage layer a potential bottleneck in deep learning training/inference.
Offline inference job usually consumes and produces tens of tera-bytes data while running more than 10 hours.
For a large-scale job, it usually causes high IO pressure, increase job failure rate, and bring many challenges for system stability.
We adopt alluxio which acts as an intermediate storage tier between the compute tier and cloud storage to optimize IO throughput of deep learning inference job.
For the production workload, the performance improves 18% and we seldom see job failure because of storage issue.
ALLUXIO DAY III 2021
April 27, 2021
Increasingly powerful compute accelerators and large training dataset have made the storage layer a potential bottleneck in deep learning training/inference.
Offline inference job usually consumes and produces tens of tera-bytes data while running more than 10 hours.
For a large-scale job, it usually causes high IO pressure, increase job failure rate, and bring many challenges for system stability.
We adopt alluxio which acts as an intermediate storage tier between the compute tier and cloud storage to optimize IO throughput of deep learning inference job.
For the production workload, the performance improves 18% and we seldom see job failure because of storage issue.
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.