A distributed compute framework for scaling AI workloads. Created and developed by @anyscalecompute.
Joined August 2019
- Tweets2.1K
- Following2
- Followers11.7K
- Likes2K
ray retweeted
@Notion is scaling AI search for millions of users.
Mickey Liu shares how moving from Spark to Ray Data on Anyscale helps generate embeddings across terabytes of data, with CPU and GPU workloads in one pipeline.
Watch: na2.hubs.ly/H08TrWn0
ray retweeted
Anyscale is at #KubeCon in Salt Lake City this week. Find us at booth #P317, join KubeRay 101 on Nov 11, and catch Rueian Huang on The Scheduling Stack panel on Nov 12.
na2.hubs.ly/H08TtmM0
ray retweeted
📍Nov 17-20 we're at Microsoft Ignite, Moscone SF. Booth 5389L.
The next wave of foundation models is coming from enterprises. Every company needs at least one FM it controls: open model, post-trained on your data, on your infra.
Session Nov 18 (THR900). Learn more: na2.hubs.ly/H08V4YK0
ray retweeted
Big news at KubeCon Americas.
KubeRay is joining the @CloudNativeFdn
Keep your KubeRay stack and add the Anyscale Control Plane in minutes. Find us at #KubeCon booth #P317.
na2.hubs.ly/H08TY0t0
ray retweeted
@Mind_Robotics trains robot foundation models on egocentric video captured on automotive production lines.
On Nov 19, they'll present what works and what doesn't.
Meetup is packed, but we saved a few seats for last-minute registrations: na2.hubs.ly/H08RLpt0
ray retweeted
Physical AI Night is back on Nov 17 in SF!
We’re teaming up with @Foxglove & @nebiusai to hear from leaders at @BedrockRobotics, @DexterityInc, @wayve_ai & @ThirdWaveAi on the frontier of robotics and autonomy.
Save your seat: na2.hubs.ly/H08Qb7p0
ray retweeted
For post-training AI for science, a genome doesn't fit in a 16k context window. So we went to 1M.
To solve this, the SkyRL team at Anyscale worked with @LilaSciences to post-train GLM-5.3-Flash at 1M tokens per sequence on 64 B200 GPUs, using a fully open-source stack.
Learn more: na2.hubs.ly/H08Q9vw0
ray retweeted
In this session, @erictang000, Software Engineer at Anyscale, will walk through:
- Intro to LLM post-training
- Intro to RLVR (with GRPO)
- Recent development of LLM post-training:
- On-policy distillation (OPD) and Multi-teacher On-policy distillation (MOPD)
- Case study: Post-training a terminal agent
- Building evals
- Building custom environments
- RL systems
- Demo walkthrough
- Intro to SkyRL
- Full Async RL with SkyRL
- Performance monitoring and tuning for RL rollouts
Register: na2.hubs.ly/H08wNDx0
ray retweeted
Lots of great topics. This covers
- SkyRL at Mercor
- What’s new in SkyRL
- 1M-context training
- Token-in token-out
- Ray Sandboxes
- Harbor
- Terminal-Bench
We're hosting an RL / data / SkyRL meetup with @mercor and @harborframework next Thursday in SF. Come join us!
luma.com/rl-data
Learn about RL and data at this meetup!
We're hosting an RL / data / SkyRL meetup with @mercor and @harborframework next Thursday in SF. Come join us!
luma.com/rl-data
ray retweeted
Post-training is now a core workload for AI labs, data providers, and enterprise leveraging proprietary data to build custom models.
Learn from the practictioners doing it at @harborframework, @NovaSkyAI and @mercor
Doors open 5:30 PM 🍕 na2.hubs.ly/H08MNd-0
ray retweeted
A failed GPU job used to be a quick kubectl logs. Then a second team joins, and the cluster that had the answer is already gone, GPUs still billing on the rerun.
Six field-tested KubeRay practices for that moment, and where KubeRay Connect picks up:
na2.hubs.ly/H08M5LS0
ray retweeted
Today, PyTorch Foundation project, Ray is announcing the Special Interest Group (SIG) for Ray on Accelerators.
Ray already supports more than a dozen accelerator types, from NVIDIA and AMD GPUs to TPUs and Trainium. The SIG brings hardware vendors, cloud providers, and the Ray community together around a vendor-neutral roadmap focused on:
· Development and testing infrastructure on real hardware
· Hardware-aware scheduling that understands topology
· Direct device-to-device transport with Ray Direct Transport
Join us on the Ray Slack channel, at community meetings, or on GitHub.
Learn more here: bit.ly/47V3QCZ
@raydistributed
ray retweeted
Anyscale is at #KubeCon in Salt Lake City next week.
Find us at booth #P317, join KubeRay 101 on Nov 11. na2.hubs.ly/H08M3Q90
Day-0 RL support for Le Chonk in SkyRL.
We have Day-0 RL support for Le Chonk on SkyRL, with LoRA training on FSDP or Megatron, courtesy of @AoShen98662303 and the @vllm_project team!
4
676
ray retweeted
If you want to learn more about reinforcement learning for post-training, check out this webinar. We're covering
- How the underlying algorithms work (e.g., RLVR with GRPO)
- Recent research advances (on-policy distillation (OPD), multi-teacher on-policy distillation (MOPD))
- Details of post-training a terminal agent (building evals, building custom environments, RL systems)
- SkyRL deep dive (async RL, performance monitoring, tuning for RL rollouts)
anyscale.com/events/2026/11/…
If you want to learn about SkyRL, this is the place to do it.
If you want to learn more about reinforcement learning for post-training, check out this webinar. We're covering
- How the underlying algorithms work (e.g., RLVR with GRPO)
- Recent research advances (on-policy distillation (OPD), multi-teacher on-policy distillation (MOPD))
- Details of post-training a terminal agent (building evals, building custom environments, RL systems)
- SkyRL deep dive (async RL, performance monitoring, tuning for RL rollouts)
anyscale.com/events/2026/11/…
ray retweeted
Last chance to sign up for next week's webinar!
Don't miss the opportunity to learn about large scale inference:
- How inference engine optimizations work (KV cache, paged attention, prefix caching, continuous batching)
- Ray Serve along with @vllm_project and @sgl_project
- Custom LLMs & integrating them with agents
- Deployment best practices (scaling, routing, composition)
- How to compare GPU & token costs
Register: na2.hubs.ly/H08jzDW0
ray retweeted
Most speech AI is built to understand intent.
@StimulerApp needs to know exactly what was said, phoneme by phoneme, to coach non-native English speakers in real time.
Custom models. Four independent stacks. No Kubernetes. Three ML engineers.
Full case study: na2.hubs.ly/H08FDlm0
