A distributed compute framework for scaling AI workloads. Created and developed by @anyscalecompute.
- Tweets 1,958
- Following 2
- Followers 11,414
- Likes 1,895
Distributed fine-tuning LLM is more cost effective than fine-tuning on a single instance! Check out the blog post on how to fine-tune and serve LLM simply, cost effectively using Ray + DeepSpeed and 🤗
hubs.ly/Q01K-BLT0
3
67
288
49,920
Ray is a powerful ML framework, but with great power comes massive documentation. How can we make it more accessible?
Now, using @langchain and Ray, we can build and deploy a doc search engine in about 100 lines of code -- with a self-hosted LLM! 1/n
piped.video/watch?v=v7a8SR-s…
7
52
275
63,244
Announcing a new Ray + 🤗 @huggingface integration!
RAG is a new NLP model that uses external documents to augment its knowledge. We’ve integrated Ray with RAG:
- 🚄Speeding up retrieval calls by 2x
- 💫Improving the scalability of fine tuning
Blog: medium.com/distributed-compu…
2
47
173
We're releasing RaySGD, a pytorch library that makes distributed training cheap and simple!
Features:
- fp16 training support
- elastic training (automatic fault tolerance)
- Integrated distributed HPO (w/ RayTune)
- intuitive and pytorch-friendly APIs
medium.com/distributed-compu…
1
54
170
Announcing Ray 2.4.0: Infrastructure for LLM training, tuning, inference, and serving.
🧠 LLM features
💽 Ray data for ease of use & stability
📊 Serve observability
🤖 RLlib’s module for custom reinforcement learning
🏢Ray scalability for large clusters
hubs.ly/Q01MYBLr0
36
161
22,726
ML serving infra has evolved, and there are 3 key requirements
- Framework agnostic (@TensorFlow, @PyTorch, pure Python, ...)
- Pure Python (intuitive for developers)
- Out of the box scalability
Why? How does this relate to Ray and @huggingface? 🤗 👇
medium.com/distributed-compu…
2
47
142
@BytedanceTalk, the company behind TikTok, uses Ray for fast & cheap offline inference with multi-modal #LLMs. They generate embeddings for a staggering 200 TB of image and text data using a model with >10B parameters.
anyscale.com/blog/how-byteda…
🧵 Thread below 👇
2
29
113
60,710
You can now tune your @huggingface transformer Trainer with RayTune (tune.io/) in 1 line of code!
⚡️Access Bayesian Optimization, Population-based Training to superpower your model
🧙♂️Use Multi-GPU and Multi-node support
Blog post: anyscale.com/blog/hyperparam…
29
117
Ray 1.0 is up on Github and PyPI (w/ new beautiful docs - docs.ray.io/en/latest/index.…)!
🎉This is a huge and important release, with many new APIs and tons of new committers!
🔖 Read about Ray 1.0 on our blog post (anyscale.com/blog/announcing…)
55
107
🎉 Say hello to Ray Lightning — a faster and simpler path to multi-node distributed training for @PyTorchLightnin⚡️.
Change 1 line to scale your PyTorch Lightning training to a multi-node GPU cluster. Give it a try and let us know what you think!
anyscale.com/blog/introducin…
25
99
Part 2 of our Ray + LangChain Series is ready, in this part we’ll show you how to turbocharge generation of embeddings.
See the video(9 minutes) at hubs.ly/Q01Np5sh0 and blog post at hubs.ly/Q01Np8090
1
32
96
18,882
ByteScale is a new LLM training framework
- Evaluated 7B to 141B param models
- 256K to 2048K context lengths
- 12,000 GPUs
- Optimized for mixed long and short sequences
The crux of it is a much more dynamic parallelism strategy (as opposed to a static mesh) to account for heterogeneity in sequence length. They call this strategy Hybrid Data Parallelism (HDP), which combines regular data parallelism with context parallelism in a dynamic manner.
Their data loading strategy is very network and CPU-memory intensive and requires global coordination across workers (as opposed to each worker doing its own thing). They use Ray actor for this coordination. There are
- Servers to fetch and preprocess raw data from HDFS and generate metadata
- A scheduler to collect global metadata from all servers, figure out the the loading plan, and broadcast the plan to clients
- Clients (on GPUs), which read the partial data from servers based on the loading plan
2
19
93
18,448
vLLM + Ray is a powerful combo for post-training.
OpenRLHF is a pioneering framework to use vLLM for RLHF, driving many design and implementation of vLLM's features for RLHF, making vLLM a popular choice for many RLHF frameworks. Learn more about the story at blog.vllm.ai/2025/04/23/open…
10
92
8,613
hyperparameter tuning for #NLProc is often overlooked, but by using @huggingface transformers + tuning techniques such as PBT, you can increase model accuracy by up to 5% on certain fine-tuning tasks
*without increasing your compute budget*!
🔖 read it:
medium.com/@amog_97444/c4e32…
24
85
The team @MetaAI has done a tremendous amount to move the field forward with the Llama models. We're thrilled to collaborate to help grow the Llama ecosystem.
anyscale.com/blog/anyscale-a…
2
18
81
77,568
JAX is a system for high-performance machine learning research and numerical computing.
At #RaySummit, @GoogleAI's @SingularMattrix will show how JAX is used in #neuralnet training, probabilistic programming & more.
Register to join live or on-demand
bit.ly/3vUpv9x
18
74
Data scientist != infra engineer.
Thanks @marksaroufim for joining our Ray Meetup last week and sharing how to make it easier to train large-scale #ML jobs in #opensource.
If you missed it, you can watch the recording here: ow.ly/SqcS50IewPm
1
13
84
excited to see Ray Tune integrated into the awesome 🤗@huggingface Transformers!
Hyperparameter search with optuna or Ray Tune is now fully integrated in Trainer (support for TF coming soon!) Tutorials coming soon but in the meantime the docs are a good way to get started with it huggingface.co/transformers/…
11
68
ICYM our blogs on Ray and Generative AI. We have a three-part series on how to use Ray to productionize common generative AI model workloads. Here are parts 1 and 2:
👉 hubs.ly/Q01JcZyd0
👉hubs.ly/Q01JcYmq0
#Ray for #GenerativeAI #workloads
11
51
50,135
🎉 Announcing Ray Serve and Anyscale Services general availability!
Teams at @LinkedIn, @Samsara, @AntGroup + many more have been using Ray to serve LLMs & multi-modal applications in a flexible, performant and scalable way.
Read more about the GA release and how companies have been using it for both:
- 🛠️ development (python API + local testing, model composition & multiplexing, heterogeneous cluster support, etc.)
- 🚀 production (high availability, observability tools, autoscaling, canary rollouts, etc.)
anyscale.com/blog/tackling-t…
1
14
53
10,865
Ray operates at two levels: Ray Core, which scales Python functions and classes with tasks and actors, and its libraries, offering easy-to-use abstractions tailored for ML workloads. #Ray #ML #DistributedComputing
8
49
16,600
We don't hear the term *exabyte* too frequently. This is an impressive use case.
aws.amazon.com/blogs/opensou…
11
51
25,560
We pretrained a stable diffusion model on 2 billion images for under $40K. Here's what we learned.
anyscale.com/blog/scalable-a…
1
15
48
24,114
Cloud TPUs from @googlecloud are one of the most cost-effective ways to train and serve LLMs.
In 2.7, Ray finally will support TPUs natively -- Ray enables a more intuitive TPU developer experience, allowing you to train and serve on massive TPU pods with ease.
Learn more at Ray Summit raysummit.anyscale.com/agend…
9
43
8,774
Deep RL has become fairly capable of optimizing reward; however, how do you choose the reward function to be optimized?
@pabbeel will discuss some recent progress in this area in his #RaySummit talk "Human-in-the-Loop Reinforcement Learning"
Register: bit.ly/3ij3OMw
7
38
🚀 Announcing the Ray Distributed Debugger! 🚀
An integrated debugging experience within VSCode.
1⃣ Set breakpoints to pause tasks and inspect variables.
2⃣ Post-mortem debugging: Analyze state after an error.
More: anyscale.com/blog/ray-distri…
piped.video/watch?v=EiGHHUXL…
10
37
12,208
Offline Batch Inference: Comparing Ray, Apache Spark & SageMaker.
Image classification benchmarks show that #Ray outperforms while linearly scaling to TB-level data sizes 💽
📈 SageMaker Batch Transform by 17x
📊 Apache Spark by 2x and 3x
hubs.ly/Q01NHV2K0
8
37
9,040
What enables Ray to be so much faster than Python multiprocessing? A combination of efficient handling of numerical data through @ApacheArrow and a set of abstractions more appropriate for building stateful services/actors. towardsdatascience.com/10x-f…
1
12
32
🎉🍾🥳 Ray 1.3 is out!
Featuring:
* Published scalability limits (github.com/ray-project/ray/t…)
* Ray Client enabled by default
* Object spilling is now turned on by default.
* Faster autoscaling for Ray Tune
* R2D2 @PyTorch and TF implementation for RLlib
github.com/ray-project/ray/r…
8
27
With Ray 2.11.0, we switched to weekly releases (previously every 6 weeks)!
This is a huge change and will get features and fixes to users faster.
This has been a big investment in our overall velocity.
github.com/ray-project/ray/r…
3
31
5,965
🔥 Modin (github.com/modin-project/mod…) is a popular library that can scale your pandas workflows by changing one line of code -- using Ray!
Learn how below:
medium.com/distributed-compu…
8
27
Ray is emerging as a standard for AI workloads, powering AI at companies like OpenAI, Uber, and Netflix. What sets Ray apart is its rich ecosystem of libraries tailored for various distributed computing tasks across the AI lifecycle.
docs.ray.io/en/latest/index.…
8
29
6,502
Growing demand for applications & HW specialization create huge burdens for learning systems at the center of intelligent applications today.
At #RaySummit, see how @tqchenml addresses these challenges using the @XGBoostProject @ApacheTVM systems he built bit.ly/3b28Atg
8
28
🎉 Microsoft Researchers have developed FLAML (Fast Lightweight AutoML) which can now utilize Ray Tune for distributed hyperparameter tuning to scale up FLAML's resource-efficient & easily parallelizable algorithms across a cluster! 🎉
Learn more: anyscale.com/blog/fast-autom…
1
5
30
Use gang-scheduling on Ray Clusters on #Kubernetes w/ #KubeRay & Multi-Cluster-App-Dispatcher (MCAD) to scale training #GLUE workloads
👉 Easy MCAD + KubeRay integration to scale Ray Clusters on #k8s
👉 Accelerate fine-tune #NLU tasks w/ multiple GPUs
anyscale.com/blog/gang-sched…
13
26
See how we optimized large-scale ML training in Part 3 of our Stable Diffusion series! We used Ray Train, Ray Data, and PyTorch Lightning to train on 2B images with fault tolerance, data streaming, and advanced strategies like FSDP and DDP.
Read more: anyscale.com/blog/we-pre-tra…
1
10
30
9,803
Blog: anyscale.com/blog/llm-open-s…
@langchain provides an amazing suite of tools for everything around LLMs.
There are tools (chains) for prompting, indexing, generating and summarizing text. While an amazing tool, using Ray with it can make LangChain even more powerful. 2/n
1
2
29
2,319
✨Ray is becoming a critical component for the next generation of ML platforms!
Check out this recent blog post about how @Uber is leveraging Ray for elastic deep learning with Horovod to enable their rapidly growing usage of deep learning:
eng.uber.com/horovod-ray/?ut…
1
25
Just dropped Ray 2.21.0, our 4th weekly release (since we started doing weekly releases).
- Ray Data @lancedb connector
- Ray Data improved retries
- Ray Serve improved batching
- Many fixes across libraries
- Improved dashboard
Tons of contributors over the past week 👇
4
25
4,765
Imagine if your random forest classifier training/tuning was 30x faster while getting 5% more accurate. Wouldn't that be awesome?
Today, by leveraging the RAPIDS library with Ray Tune, you can do that. See how in exciting new post: medium.com/rapids-ai/30x-fas…
#GTC2020 #RayTune
With @rapidsai and @raydistributed #RayTune, you can now tune Random Forest Classifiers 30x faster -- while getting a 5% accuracy boost. See how. nvda.ws/2WCsqou
12
27
Exciting talk from @dariogila with @IBM on the future of quantum computing, and how @raydistributed could be the key for its success.
8
27
0.8.6 is out!
- Support for Windows (alpha)!
- Releasing Ray Serve, a scalable model-serving library! Check out a tutorial for serving @PyTorch models: docs.ray.io/en/master/serve/…
- Ray Dashboard now supports GPU monitoring!
And more!
Release notes: github.com/ray-project/ray/r…
6
27
As technology has advanced, ML architectures have evolved. One way to see it is in terms of generations:
- 1st gen involved "fixed function" pipelines
- 2nd gen involved programmability within the pipeline
What will be the next gen of ML architectures?
bit.ly/2XyG9zR
7
28
🎉🍾🥳 Ray 1.5 is out!
Featuring:
- Ray Datasets now in alpha
- LightGBM on Ray in beta
- The Ray cluster launcher now has support for launching clusters on Aliyun
- RLlib added an improved "input API" for customizing offline datasets
Learn more ⬇️
github.com/ray-project/ray/r…
1
2
27
Announcing Ray 2.5 release features:
👉 Support #LLMs training with Ray Train
👉 Serve #LLMs with Ray Serve
👉 Multi-GPU learner stack in #RLlib for cost efficiency & scalable RL-agent training
👉 Performant & improved approach to batch inference at scale
hubs.ly/Q01TjbM00
1
9
25
2,506
Check out the new @MLflow and @raydistributed integrations for tuning models, tracking experiments, and deploying models.
medium.com/distributed-compu…
1
1
27
First sessions for #RaySummit program are up! Join the annual gathering of the global @raydistributed community for the latest in distributed computing.
Speakers include @TravisAddair @eric_brewer @tqchenml @slbird @dawnsongtweets & more
➡️raysummit.org
12
25
"A Step-by-Step Guide to Scaling Your First Python Application in the Cloud" by Bill Chambers link.medium.com/W0Yj2hbNg6. You'll learn how to install Ray, create an app, test on your local machine, spin up a Ray cluster in the cloud, deploy your app, ... and more!
5
24
Streaming distributed execution across CPUs and GPUs: Learn how Ray Data streaming works and how to use it for your own ML pipelines.
hubs.ly/Q01Px5WS0
1
4
23
2,046
Just dropped Ray 2.24.0 🥂🥳
🎗️ Tons of new work on observability, particularly around machine failures. Why did a node die (failure, scaling down, spot preemption, etc).
🔥 Critical bug fixes across Ray core and the Ray dashboard.
🎂 New features in Ray Data and Ray Serve.
8
26
5,328
Distributed libraries allow improved performance by exploiting the full bandwidth of distributed memory, and giving greater programmability. But how does that actually work? What does the code look like?
Learn more ⬇️
bit.ly/3o6a6l8
6
25
🎉 New Introductory Tutorial on Reinforcement Learning (RL) with OpenAI Gym, RLlib, and Google Colab! 🎉
bit.ly/2Wlnx5W
The tutorial explores:
- What is RL
- The OpenAI Gym CartPole Environment
- The Role of Agents in RL & how to train them using RLlib
7
24
AI for protein design.
Amazing work by @DreamFoldAI.
FoldFlow-2 is a generative model for protein structure, which is important for protein design.
Trained on @anyscalecompute.
5
25
2,862
ML serving is broken - Ray Serve can fix it! Thread (1/n)
🙁Wrapping models in Flask doesn’t scale
🙁TorchServe, TFServing requires setting up a traditional web server
😊 Ray Serve lets you deploy your ML models with a simple Python interface!
medium.com/distributed-compu…
1
5
23
💥🎉 Ray version 1.9 is here!
Featuring:
✅ Ray Train is now in beta!
✅ Ray Datasets now supports groupby and aggregations!
✅ Ray Docker images for multiple CUDA versions are now provided!
bit.ly/3pzpIgr
1
2
22
🎉🍾🥳 Ray 1.4 is out!
Highlights include:
- Ray Serve has a new deployment centric API!
- Ray now has support for namespaces. (Docs: docs.ray.io/en/master/namesp…)
- RLlib now has multi-GPU support for PyTorch models!
Learn more ⬇️
github.com/ray-project/ray/r…
6
22
🎉 New Tutorial on Serverless Kafka Stream Processing with Ray!
Featuring:
- Ray Clusters that autoscale to meet the demands of a stream processing job
- How Ray can be paired with @apachekafka
Learn more ⬇️
anyscale.com/blog/serverless…
6
22
💥🎉 Ray version 1.8 is here!
Featuring:
✅ Ray SGD has been renamed to Ray Train
✅ Ray Datasets, now beta, has a new integration with Ray Train for scalable ML ingest
✅ Experimental support for Ray on Apple Silicon (M1 Macs)
bit.ly/3k3eIGw
5
22
Uber built a unified ML platform that abstracts away infra complexity — letting teams run Ray jobs without worrying about clusters or resource placement.
@raydistributed + @kubernetesio handle orchestration and scaling across @Uber's fleet. 🤝
Full setup breakdown 👇
uber.com/blog/ubers-journey-…
4
24
8,021
Ray 2.3.0 Released with:
⭐️ Observability enhancements
⭐️ Ray Dataset Streaming
⭐️Boost in Ray core performance
⭐️Gym/Gymnasium library in #RLlib
⭐️ Support ARM & Python 3.11
⭐️ Support multiple applications in Ray Serve (developer preview)
anyscale.com/blog/announcing…
3
23
1,454
Ray continues to enable #ML teams innovate at scale & unleash new use cases. @Spotify shares how #Ray helps #ML practitioners innovate & how they built ML platform atop Ray.
"Our goal for Spotify’s ML Platform has always been to create a seamless user experience for ML practitioners who want to take an ML application from development to production..."
And so, we introduced @raydistributed to our @Spotify ecosystem. engineering.atspotify.com/20…
5
22
2,693
🙌🙌 With the v3.0 release, you can use Ray to train @spacy_io on one or more remote machines, potentially speeding up your training process.
explosion.ai/blog/spacy-v3-n…
9
19
Co-creator of @PyTorch at Meta AI @soumithchintala shares how various project co-exist with @raydistributed at #raysummit.
4
22
Surprisingly, most popular key-value stores don't support shared-memory! The Plasma Store, part of @ApacheArrow, does. In conjunction with Arrow’s data layout, this enables super fast sharing of data between multiple processes on the same machine. ray-project.github.io/2017/0…
1
13
23
Ray and Uber have a long history, including
1. uber.com/blog/horovod-ray/
2. uber.com/blog/elastic-xgboos…
3. uber.com/blog/from-predictiv…
This is a fantastic read on Uber's AI 8 year AI journey. From (1) predictive ML on tabular data to (2) adopting deep learning to (3) venturing into generative AI.
It's amazing to see that @raydistributed has played a role in enabling deep learning and LLM training at Uber.
5
22
3,437
Announcing a collaboration between PyCaret + Ray!
🔥PyCaret (pycaret.org/) is a popular low-code ML library in Python. A new contributed blog shows how #PyCaret integrated Ray's tune-sklearn (github.com/ray-project/tune-…) to simplify model tuning!
medium.com/distributed-compu…
4
21
At #RaySummit, @vanpelt will discuss the @wandb tool Tables + new Artifacts features that let you visualize & query datasets & model evaluations at the example level as well as integrate with Ray.
Register: bit.ly/3fLnRBA
6
20
🎉New blog post on the most popular RL talks from Ray Summit 2021!
Including:
- 24x Speedup for RL (Raoul Khouri)
- Orchestrating Robotics Operations with SageMaker + RLlib (@SahikaGenc)
- Offline RL with RLlib (@edilmop)
- Neural MMO (@jsuarez)
bit.ly/3hXrF3I
5
16
New blog post, "Scaling Python Asyncio with Ray" by Simon Mo link.medium.com/GTWU6LxWv4
12
22
Awesome turnout for @anyscalecompute @CodyHaoYu presentation at the @vllm_project meetup—nearly 300 people joined to hear about the vLLM roadmap and our team's release of new LLM APIs in Ray Data and Ray Serve.🙌
The new batch inference APIs seamlessly integrate vLLM, improving both speed and scalability.
See the APIs here:
Ray Data + LLMs-docs.ray.io/en/master/data/w…
Ray Serve for LLMs- docs.ray.io/en/master/serve/…
6
22
3,452
Incredible meetup last night. Thank you to @netflix for hosting! Great talks from
- Lingyi Liu on Netflix's ML platform
- Pablo Delgado on multimodal data curation at Netflix
- Lei Xu on LanceDB's multimodal lakehouse
- Richard Liaw on Ray Data for AI data processing
6
22
4,434
Ray 1.2 is up on GitHub and PyPI (github.com/ray-project/ray/r…)!
🎉 This is an important release with many new APIs and tons of new committers.
Some highlights 👇
2
4
22
The Ray 2.6.1 released with :
🎏 Streaming responses in Serve for real-time capabilities
🎏 📀🏃♀️Ray Data streaming integration w/Train
🏃♀️☁️Distributed Training & Tuning sync with cloud storage persistence
🤖 Alpha release of the Multi-GPU Learner API
📙 Ray Gallery examples
👇
1
3
22
6,938
ByteDance has recently shipped some very impressive models:
- OmniHuman-1 for high quality deepfake videos
- Seed-Music for music generation
Today they hosted an incredible meetup and went into detail on how they use Ray for their (1) audio processing, (2) video pipelines, and (3) RLHF.
1
3
21
4,191
In Ray 1.0.1, we're releasing Population-based Bandits (PB2), a new method for tuning neural networks published in #NeurIPS2020 by @jparkerholder and @nguyentienvu!
🚀 PB2 can perform up to 6x more efficiently than methods like Hyperband, PBT.
🔖 Read: anyscale.com/blog/population…
7
20
💥🎉 Ray version 1.7 is here!
Featuring:
✅ Ray SGD v2, now alpha, introduces APIs that focus on ease of use and composability
✅ Ray Workflows is in alpha. Try it out for your large data, ML, and business workflows
✅ Major enhancements to the C++ API
bit.ly/3mH4tIm
1
5
20
Distributed C++ systems are more difficult to put into production than single machine systems due to communication, deployment, and fault tolerance issues.
The new Ray C++ API was designed to help to address these problems.
Learn more ⬇️
bit.ly/3oeWTFq
2
20
⚡️In Ray 1.2, we’re improving Ray support for distributed data processing! Featuring:
- 💿External storage support
- ✨Support for Python data processing libraries
Use @ApacheSpark , @dask_dev DataFrames alongside ML libraries on Ray like Horovod!
Blog: medium.com/distributed-compu…
4
18
🎉 Introducing Distributed XGBoost Training with Ray!
Featuring:
- Distributed training by only changing three lines of code
- Distributed hyperparameter tuning with Ray Tune
- Support for Pandas, Modin, & even Dask Dataframes!
Learn more ⬇️
bit.ly/35rjtlc
1
6
19
As part of our efforts on #observability, a novel feature: "Automatic and optimistic memory scheduling for ML workloads in Ray"
👉 minimal configuration
👉 policy-based mitigation of #OOM errors w/retriable tasks
👉 debug OOM problems w/ the monitor
anyscale.com/blog/automatic-…
1
6
21
1,573
4 common patterns of serving ML models in production are: pipeline, ensemble, business logic, & online learning.
Implementing these patterns typically involves a tradeoff between easy development and production readiness. Learn how Ray Serve changes this
bit.ly/3ipEsMi
6
20
Very impressive to see how @canva is using LLMs and image generation to transform the design world.
Canva is a leader in generative AI and modernized their AI platform with @raydistributed.
Some key challenges
- Scaling training on more GPUs and far more data.
- Unifying generative AI and non-generative models.
- Flexibility to support different clouds and accelerators.
This enabled @canva to speed up training by an order of magnitude and fully saturate GPU utilization.
anyscale.com/resources/case-…
1
5
18
9,135
🎉 Ant Group has developed Ant Ray Serving which is an online service framework based on Ray, which provides users with a Serverless platform to publish Java/Python code as online services & allows them to focus on their own business logic 🎉
Learn more: bit.ly/3yUkKxy
5
19
There’s an even divide between developers choosing a generic #Python web server such as @FastAPI and a specialized ML serving solution framework.
Check out our latest blog post for more on each option and explore why you might choose one over the other: ow.ly/tlhG50K7I1m
5
21
You can configure and Scale ML with @Hydra_Framework and Ray on AWS or local Ray clusters.
Blog Post: medium.com/distributed-compu…
5
21
🎉 Really exciting blog from @UberEng on moving distributed @XGBoostProject onto Ray along with parallel efforts to move Elastic #Horovod onto Ray!
This is a critical step towards a unified distributed compute backend for end-to-end machine learning workflows at Uber!
New on our blog today! Members of our engineering team describe how they co-developed Distributed XGBoost on Ray with the Ray team @raydistributed to tackle various production challenges of doing distributed machine learning at scale.
read more: eng.uber.com/elastic-xgboost…
3
18
After training a #MachineLearning model, the model needs to be deployed for online serving and offline processing.
At #RaySummit, @simon_mo_ will walk through the journey of deploying ML models in production and how Ray Serve was built.
Register: bit.ly/3wdvWo2
7
18
As modern hardware systems get more complex, it’s becoming more difficult to design integrated circuit implementations.
Check out the blog post from the @IBMResearch team to learn how they use AI/ML-driven chip design and Ray to solve this challenge: ow.ly/PX1Q50Jzc8O
14
19
Read this blog to learn about Composer, Cursor's latest frontier model built with Ray. For the technical deep dive, come to Ray Summit next week! cursor.com/blog/composer
2
6
21
4,653
SGLang 🤝 Ray!
We're super excited to have @ying11231 and @liin1211 talk about SGLang and its new features at Ray Summit!
They'll highlight the newest SGLang features and also talk about SGLang's integration with Ray Data LLM.
Hope to see you there!
SGLang at Ray Summit 2025 is coming!
📍 San Francisco • Nov 3–5 • Hosted by @anyscalecompute
🗓 On Nov 5, SGLang is invited to give a talk on Efficient LLM Serving
🎤 @ying11231 & @liin1211 will introduce core features, high-throughput & low-latency tricks, real-world deployment lessons, and the future roadmap.
✨ Use RaySGLang50 for 50% off!
For anyone who cares about: Distributed AI at scale, Performance & efficiency, Open-source evolution - Tag a friend who should join!
#SGLang #RaySummit2025 #RayData #DistributedAI
anyscale.com/ray-summit/2025
1
5
22
5,175
#RaySummit is almost here!
Don’t miss out on:
🌁 In-person networking in SF
🎒 3 in-depth Ray training sessions
⚙️ 40+ technical sessions and lightning talks
🎤 Speakers from @MetaAI, @Spotify, @IBM & more
...and much more!
ow.ly/RpO050KiKQZ
1
8
19
Ray has many ML integrations such as Horovod and 🤗 to data processing frameworks such as Spark, Modin, and Dask.
But what does it mean to be "integrated with Ray"? And what benefits does it provide to library developers and users?
Learn more ⬇️
bit.ly/2TvQxGh
7
19
Since it was first released Ray Tune is a leading way of scaling ML tuning. But there's a gap - experiment management & ML tracking. To close this, we're happy to announce an integration with @wandb ! Read about it here: medium.com/distributed-compu…
1
6
19
🎉 Ray 1.12 is here! This release includes the alpha of Ray AI Runtime (AIR), a new, unified experience for seamless integration across the Ray ecosystem.
📢 Shoutout to all of the community members who supported this release. Learn all about it here: ow.ly/OF3350IKnw4
1
8
17
A distributed shuffle is a data intensive-operation that usually calls for a system built specifically for that purpose.
Even though its core API contains no shuffle operations, Ray can do it in just a few lines of Python. Learn how 👇
medium.com/distributed-compu…
1
16
The performance numbers resulting from the ongoing re-architecture are impressive! Here's why: medium.com/distributed-compu…
2
20
Using @raydistributed with @scikit_learn . @AmeerHajAli shows you how. medium.com/distributed-compu…. The technique leverages Ray's implementation of joblib. He also shows performance measurements of Ray vs. other tools, Loky, Multiprocessing, and Dask.
10
18
PyTorch + vLLM + Kubernetes + Ray is a great combination.
An #OpenSource Stack for #AI Compute: @kubernetesio + @raydistributed + @pytorch + @vllm_project ➡️ This Anyscale blog post by @robertnishihara describes a snapshot of that emerging stack based on experience working with Ray users + case studies from Pinterest, Uber, Roblox, and 5 popular open source post-training frameworks hubs.la/Q03tnBZD0
5
18
1,630
Here is part 2, zooming way in on data preparation (along with runnable code).
anyscale.com/blog/processing…
2
6
18
8,904