Some stats for this release
🍒 Zero cherrypicks (despite releasing on all platforms)
💇‍♀️ Released 2 days after branch cut

Sounds easy but it really isn't 😅

Weekly releases mean that we get features & fixes to our users faster.

We released Ray 2.20 this week! (our third weekly release)

You'll notice the numbering has changed (incrementing from 2.12.0 -> 2.20.0).

The bump indicates a release on all platforms, whereas our weekly release cadence includes a subset of platforms.

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 ray has played a role in enabling deep learning and LLM training at Uber.

Ray and Uber have a long history, including

1. uber.com/blog/horovod-r…
2. uber.com/blog/elastic-x…
3. uber.com/blog/from-pred…

One of my favorite talks from Ray Summit last year was Brian McClendon on how Niantic built their AI infra with Ray.

'Reduced latency by 75%, cost by 60-70%, and lines of code by 85%.'

📣Calling all ray users: Share your story with the Ray Community at #RaySummit 2024!

Call for proposals is open through Thursday, May 9th! Submit here: hubs.ly/Q02vL9w00.

#ray #ray summit #AI #LLM

Speak at Ray Summit 2024!

Welcome Anyscale! The seamless integration of MongoDB Atlas data platform and vector search with the Anyscale AI platform empowers developers to build, fine-tune, and deploy enterprise grade #GenAI and other #AI apps, hassle-free.

Try out Anyscale plus MongoDB.

Some of our earliest use cases were for reinforcement learning. Awesome to see this work from Character.AI.

Speak at Ray Summit!

Last year we heard from
OpenAI
Adobe
Niantic
Airbnb
Pinterest
Uber
LinkedIn
LangChain
LlamaIndex 🦙
Meta
NVIDIA

and many many others.

Ray Summit 2024 call for talks is now open!

anyscale.com/blog/ray-summi…

Congratulations to everyone who helped make this happen! This is a big investment in our overall velocity.

It's hard to over invest in developer productivity.

Weekly releases is a big deal. The pace can only increase 🔥

Instruct variants of Llama-3 models are now available on Anyscale Endpoints. Give it a spin✨

For large scale documents, generating vector embeddings offline can save significant GPU resources. Check out how to efficiently generate embeddings and load vectors for search with Ray Data and Milvus. Milvus Zilliz ray

Read: hubs.ly/Q02sWLST0