Speed and efficiency are two requirements for the underlying infrastructure for machine learning model development. Data access can bottleneck end-to-end machine learning pipelines as training data volume grows and when large model files are more commonly used for serving. For instance, data loading can constitute nearly 80% of the total model training time, resulting in less than 30% GPU utilization. Also, loading large model files for deployment to production can be slow because of slow network or storage read operations. These challenges are prevalent when using popular frameworks like PyTorch, Ray, or HuggingFace, paired with cloud object storage solutions like S3 or GCS, or downloading models from the HuggingFace model hub.
In this presentation, Lu and Siyuan will offer comprehensive insights into improving speed and GPU utilization for model training and serving. You will learn:
- The data loading challenges hindering GPU utilization
- The reference architecture for running PyTorch and Ray jobs while reading data from S3, with benchmark results of training ResNet50 and BERT
- Real-world examples of boosting model performance and GPU utilization through optimized data access
Speed and efficiency are two requirements for the underlying infrastructure for machine learning model development. Data access can bottleneck end-to-end machine learning pipelines as training data volume grows and when large model files are more commonly used for serving. For instance, data loading can constitute nearly 80% of the total model training time, resulting in less than 30% GPU utilization. Also, loading large model files for deployment to production can be slow because of slow network or storage read operations. These challenges are prevalent when using popular frameworks like PyTorch, Ray, or HuggingFace, paired with cloud object storage solutions like S3 or GCS, or downloading models from the HuggingFace model hub.
In this presentation, Lu and Siyuan will offer comprehensive insights into improving speed and GPU utilization for model training and serving. You will learn:
- The data loading challenges hindering GPU utilization
- The reference architecture for running PyTorch and Ray jobs while reading data from S3, with benchmark results of training ResNet50 and BERT
- Real-world examples of boosting model performance and GPU utilization through optimized data access
Video:
Presentation slides:
Videos:
Presentation Slides:
Complete the form below to access the full overview:
.png)
Videos
Fireworks AI is a leading inference cloud provider for Generative AI, powering real-time inference and fine-tuning services for customers' applications that require minimal latency, high throughput, and high concurrency. Their GPU infrastructure spans 10+ clouds and 15+ regions, serving enterprises and developers deploying production AI workloads at scale.
With model sizes reaching 70GB+, Fireworks AI faced critical challenges: eliminating cold start delays, managing highly concurrent model downloads across GPU clusters, reducing tens of thousands in annual cloud egress costs, and automating manual pipeline management that consumed 4+ hours weekly. They chose Alluxio as their solution to scale with their hyper-growth without requiring dedicated infrastructure resources.
In this tech talk, Akram Bawayah, Software Engineer at Fireworks AI, and Bin Fan, VP of Technology at Alluxio, share how Fireworks AI uses Alluxio to power their multi-cloud inference infrastructure.
They discuss:
- How Fireworks AI uses Alluxio in its high-performance model distribution system to deliver fast, reliable inference across multiple clouds
- How implementing Alluxio distributed caching achieved 1TB/s+ model deployment throughput, reducing model loading from hours to minutes while significantly cutting cloud egress costs
- How to simplify infrastructure operations and seamlessly scale model distribution across multi-cloud GPU environments
