@aiatmeta - AI Dynamics

vLLM Kernel Optimizations Boost GB200 Inference Performance

By

–

04 February 2026 21h28

Impressive deep dive! It’s great to see the vLLM team maximizing the GB200’s potential. These kinds of kernel-level optimizations are exactly why the PyTorch ecosystem continues to be the foundation for next-gen inference performance.

→ View original post on X — @aiatmeta

4 February 2026

Segment Anything Models Advance Flood Monitoring and Disaster Response

By

@aiatmeta

–

22 December 2025 17h07

Our Segment Anything Models are helping advance flood monitoring and disaster response.

See how @USRAedu and @USGS have fine-tuned SAM to automate a key bottleneck in real-time river mapping, enabling faster, scalable, and more cost-effective disaster preparedness:… pic.twitter.com/kSHkaOKlhe
— AI at Meta (@AIatMeta) 22 décembre 2025

Our Segment Anything Models are helping advance flood monitoring and disaster response. See how @USRAedu and @USGS have fine-tuned SAM to automate a key bottleneck in real-time river mapping, enabling faster, scalable, and more cost-effective disaster preparedness:

→ View original post on X — @aiatmeta

22 December 2025

Collaborative Progress Accelerates Innovation and Advancement

By

@aiatmeta

–

18 December 2025 19h53

The frontier moves faster when we all pitch in

→ View original post on X — @aiatmeta

18 December 2025

Meta Open-Sources PE-AV for Advanced Audio Separation Technology

By

@aiatmeta

–

18 December 2025 17h58

We’re open-sourcing Perception Encoder Audiovisual (PE-AV), the technical engine that helps drive SAM Audio’s state-of-the-art audio separation. Built on our Perception Encoder model from earlier this year, PE-AV integrates audio with visual perception, achieving

→ View original post on X — @aiatmeta

18 December 2025