In the most recent Lightning Thunder Session, Luca and Tom look at doing DDP on the sharded weights from FSDP in a 2d device mesh.

Watch the full episode here ➡️ youtu.be/iIMA9_ippEw

Last month, we hosted a Robotics Meetups in New York City featuring talks from 2 industry experts:

✨ Lerrel Pinto, NYU Assistant Prof. of Computer Science
✨ Shagun Majotra, Mechatronics and Robotics Engineering Masters Student

This clip explores constructivism's relationship to building robots:

Lightning AI can cut deployment costs by 15% with fast and measurable cold-starts:

✅ Enterprise-grade reliability
✅ Scale to zero to save money
✅ Move between multiple clouds seamlessly

Check it out 👉 lightning.ai/lightning-a...

Learn how to optimize costs with this Lightning Studio template featuring 3 hands-on notebooks using NVIDIA cuML:

⚡️ Lightning-fast classification
📊 Scalable clustering
🔍 Effortless dimensionality reduction

Try it now 💥 go.lightning.ai/4cKTAPs

Save money with same day reservations of in-demand GPUs.

Instantly reserve A100, H100 and H200 machines for half the price!

This week's Thunder Session had @lantiga and Tom look forward to having a joint forward and backward trace to work with.

Check out the full episode ➡️ youtu.be/NGKIDxzjNR0

In the last Thunder Session, Luca and Tom discussed state —model params + metadata needed to save/load during training - and how it gets transformed in the FSDP Transform.

Thunder Session livestream returns Friday. Watch it live here on X or youtube: www.youtube.com/live/Hn9isu...

🚀 Instantly deploy Meta’s latest flagship reasoning model, Llama 4 API, for visual question answering, image captioning, and multimodal content generation.

Easily integrate Scout and Maverick into your applications with Lightning AI Hub’s enterprise-grade API.