🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

One of our biggest advantages is talent. We get top people from all over the world who largely stay here and contribute to innovation here (especially AI). Turning away these people is a mistake.

1/5 How do you deliver fast, personalized search to 100M+ users? You need more than a good model. You need great infrastructure. Here’s how Notion scaled their search pipeline using Ray + Anyscale ↓🧵

Very excited to be working with Notion on their AI efforts.

Looking forward to speaking at the Weights & Biases conference next week! Will be talking about open-source AI infra. wandb.ai/site/resources…

This is an incredibly technical talk from Nubank on building foundation models for financial transactions. The team that did this joined Nubank via an acquisition. They proceeded to leverage all of Nubank's existing data & model infrastructure and surgically insert foundation

My friend Robert Nishihara told me a fun math problem the other day. O3 gets it wrong even after a fair amount of prodding - is this somehow harder than the olympiad problems where it gets 95% accuracy?

Ray Summit talk submissions are now open! We're very very excited to hear about your work. - How you use Ray - How you use vLLM - AI infrastructure - Multimodal data - Post-training - Agentic systems - Challenges at scale

I've already heard two companies this week say "we built out everything around text data, then we began introducing images / video and everything broke."

The AI compute software stack consists of 3 specialized layers: 🔧🔧🔧 Layer 1: Training & Inference Framework (PyTorch + vLLM) • Runs models efficiently on GPUs • Handles model optimization and model parallelism strategies • Manages accelerator memory and automatic

I get a lot of questions around what is the role of each of these layers of AI compute stack: vLLM, ray, k8s, etc. What does ray do in vLLM, what does ray do around vLLM? Why is Ray core part of post-training frameworks like vERL, etc? In this blog Robert Nishihara depicts what a

This table was a footnote at the end of the blog, but it's actually one of the most interesting points. There is an emerging stack for post-training. anyscale.com/blog/ai-comput…

An OSS stack for AI compute: Kubernetes+ Ray + PyTorch + vLLM ... from Robert Nishihara of Anyscale anyscale.com/blog/ai-comput…

Beyond pre-training, here's how I imagine most learning will work. 1. AI models / systems will maintain large collections of retrievable knowledge. This will include facts like "the capital of California is Sacramento" and tactics like "when playing Monopoly, buy a bunch of

I really enjoyed the new blog from Anyscale about open source stack for AI compute. Robert shared a great collection of examples showing how companies such as Pinterest, Uber, and Roblox integrate Kubernetes, Ray, PyTorch, and vLLM. This stack enables extensive training

Impressive work! Agentic workflows have tons and tons of design and arcitectural decisions that affect performance and quality (choices around models, embedddings, how to tokenize / chunk data, how to do retrieval, how to construct context, etc). There's a massive search space,

At a recent birthday party, we had a bunch of bingo ice breaker questions. The winner was Alex Levy.