Robert Nishihara
@robertnishihara
Followers
8K
Following
4K
Media
132
Statuses
2K
Co-founder @anyscalecompute. Co-creator of @raydistributed. Previously PhD ML at Berkeley.
Joined March 2009
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.
We're missing techniques for "training-time reasoning." Right now there's a lot of progress on inference-time reasoning, which is incredibly cool (I use o3 all the time). If I think about how I learn stuff, e.g., when reading a technical paper, it's very compute intensive. Most.
1
5
45
There are new open-source RL frameworks every week, and they largely follow this division of responsibilities.
We are seeing an emerging OSS stack for AI compute:. 🔧PyTorch + 🧠vLLM + ⚡Ray + 📦Kubernetes. 📽️ @robertnishihara breaks down how these layers work together to scale LLMs + GenAI workloads. 📰Full blog with examples from Pinterest, Uber, & Roblox:
1
2
5
RT @anyscalecompute: We are seeing an emerging OSS stack for AI compute:. 🔧PyTorch + 🧠vLLM + ⚡Ray + 📦Kubernetes. 📽️ @robertnishihara bre….
0
8
0
RT @lindavivah: 📢 Speaker submissions are now open for the vLLM track at Ray Summit 2025 in SF in November!. Btw this is the first ever Fea….
anyscale.com
Powered by Ray, Anyscale empowers AI builders to run and scale all ML and AI workloads on any cloud and on-prem.
0
2
0
Interesting open source project aiming to simulate 10s of thousands of agents to model societal interactions (using @raydistributed). "Experiments demonstrate that the framework can support simulations of 30,000 agents that are faster than the wall-clock time."
2
4
12
1
1
4
RT @lindavivah: 📣 Excited to share that I’ve joined @anyscalecompute as a Staff Developer Advocate! . This is the brilliant team behind ✨Ra….
0
8
0
DistFlow, from researchers at Shanghai Innovation Institute, looks like a cool new RL framework that builds on Ray, PyTorch, vLLM, and SGLang. @raydistributed, @PyTorch, @vllm_project, @lmsysorg.
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.
1
4
17
RT @agermanidis: Models just want to generalize. For the past years, we’ve been pushing the frontier of controllability in video, releasin….
0
31
0
I always love when companies use Ray for a large variety of different workloads.
How @klaviyo uses Ray for data processing, training, hyperparameter tuning, and model serving!.
1
3
19
RT @weights_biases: 🚀 AI workloads are exploding. @robertnishihara of @anyscalecompute shows how Kubernetes, Ray, PyTorch and vLLM snap to….
0
2
0
If you're building with @vllm_project, speak at the dedicated vLLM track at Ray Summit in November.
Last year, the creators of @vllm_project at UC Berkeley hosted a massive two-day vLLM event featuring presentations from Roblox, Uber, Apple, Intel, Alibaba, Neural Magic, IBM, Handshake, Databricks, Anyscale, and others on how they are using and optimizing vLLM. This covered
0
3
14
Everyone talks about how voice mode (once polished) will be a major UX unlock for AI, which is correct. An equally important frontier, which no one has touched yet, is AI group chats. Lots of hard product challenges to solve there, but it'll be hard to imagine AI without it once.
2
1
11
I started reading this thread and then got distracted trying to solve the math problem. It's a great problem and very enjoyable to think about. I highly encourage you to get out a sheet of paper, draw some triangles, and take a crack at it.
2/N We evaluated our models on the 2025 IMO problems under the same rules as human contestants: two 4.5 hour exam sessions, no tools or internet, reading the official problem statements, and writing natural language proofs.
0
0
8
Reinforcement learning is a big investment area for us at @anyscalecompute, and we're hiring actively for RL! If you're interested in building systems & algorithms for RL, message me.
Congratulations to my brilliant co-founder Philipp Moritz (@pcmoritz) and the legendary John Schulman, Sergey Levine, Pieter Abbeel, and Michael Jordan on their Test-of-Time Honorable Mention at ICML 2025 today!. For creating TRPO. This was done during the previous wave of
0
0
9
RT @ashugarg: Huge congrats to @pcmoritz, co-founder of @anyscalecompute for the Test-of-Time Honorable Mention at #ICML2025.
0
1
0
In large part due to Philipp's work on TRPO, reinforcement learning was one of the original motivating use cases that led us to build @raydistributed. You can see how we framed it in our early Ray paper (on page 1).
Congratulations to my brilliant co-founder Philipp Moritz (@pcmoritz) and the legendary John Schulman, Sergey Levine, Pieter Abbeel, and Michael Jordan on their Test-of-Time Honorable Mention at ICML 2025 today!. For creating TRPO. This was done during the previous wave of
1
6
18