Ray Data
Multi-Modal Data Processing Engine. Anyscale supports and further optimizes Ray Data for improved performance, reliability, and scale.
17x
faster compared to AWS SageMaker for data preprocessing
2x
faster than Apache Spark for unstructured data preprocessing
up to
4.5x
faster at running read-intensive data workloads compared to open source Ray
up to
6x
cheaper for LLM batch inference compared to AWS Bedrock and OpenAI
What is Ray Data?
Ray Data is a scalable data processing library for ML and AI workloads.
With flexible and performant APIs for distributed data processing, Ray Data enables offline batch inference and data preprocessing ingest for ML training. Built on top of Ray Core, it scales effectively to large clusters and offers scheduling support for both CPU and GPU resources. Ray Data also uses streaming execution to efficiently process large datasets and maintain high GPU utilization.
Use Cases
Offline Batch Inference
Ray Data offers an efficient and scalable solution for batch inference, consistently outperforming competitors:
- 17x faster than SageMaker Batch Transform
- 2x faster than Spark for offline image classification
Benefits
Maximize GPU and CPU Utilization
Leverage CPUs and GPUs in the same pipeline with to increase GPU utilization and decrease costs
Built-in Fault Tolerance
Ray Data automatically recovers from out-of-memory failures and spot instance preemption.
One Unified API
Work with your favorite ML frameworks and libraries, just at scale. Ray Data supports any ML framework of your choice—from PyTorch to HuggingFace to Tensorflow and more.
Any Format, Any Data Type
Ray Data supports a wide variety of formats including Parquet, images, JSON, text, CSV, and more, as well as storage solutions like Databricks and Snowflake.
Supercharge Ray Data with Anyscale
Best-In-Class Data Processing for ML/AI
Text Support
Image Support
Audio Support
Video Support
3D Data Support
Task-Specific CPU & GPU Allocation
Stateful Tasks
Native NumPy Support
Native Pandas Support
Model Parallelism Support
Nested Task Parallelism
Fast Node Launching and Autoscaling
Fractional GPU Support
Load Datasets Larger Than Cluster Memory
Improved Observability
Autoscale Workers to Zero
Job Queues
Priority Scheduling
Accelerated Execution
Data Loading / Data Ingest / Last Mile Preprocessing
| | | | |
|---|---|---|---|---|
Text Support | | | | |
Image Support | | - | | |
Audio Support | Manual | - | Manual | |
Video Support | Manual | - | Manual | |
3D Data Support | - | - | Binary | Binary |
Task-Specific CPU & GPU Allocation | - | - | | |
Stateful Tasks | - | - | | |
Native NumPy Support | - | - | | |
Native Pandas Support | | - | | |
Model Parallelism Support | - | - | | |
Nested Task Parallelism | - | - | | |
Fast Node Launching and Autoscaling | - | - | - | 60 sec |
Fractional GPU Support | - | Limited | | |
Load Datasets Larger Than Cluster Memory | - | - | | |
Improved Observability | - | - | - | |
Autoscale Workers to Zero | - | Limited | | |
Job Queues | - | - | - | |
Priority Scheduling | - | - | - | |
Accelerated Execution | - | - | - | |
Data Loading / Data Ingest / Last Mile Preprocessing | | | | |
Out-of-the-Box Templates & App Accelerators
Jumpstart your development process with custom-made templates, only available on Anyscale.
Batch Inference with LLMs
Run LLM offline inference on large scale input data with Ray Data
Computing Text Embeddings
Compute text embeddings with Ray Data and HuggingFace models.
Pre-Train Stable Diffusion
Pre-train a Stable Diffusion V2 model with Ray Train and Ray Data
“We have no ceiling on scale, and an incredible opportunity to bring AI features and value to our 170 million users.”
Greg Roodt
ML Lead, Canva
FAQs
There are a variety of Anyscale customers and Ray open source users using the Ray Data library to build advanced AI applications. A few exciting examples include:
Companies Using Ray Data for Offline Batch Inference:
- ByteDance scales offline inference with multi-modal LLMs to 200 TB on Ray Data
- Spotify’s new ML platform built on Ray Data for batch inference
- Sewer AI speeds up object detection on videos 3x using Ray Data
Companies Using Ray Data for ML Training Ingest:
The Best Option for Data Processing At Scale
Get up to 90% cost reduction on unstructured data processing with Anyscale, the smartest place to run Ray.