1. X
  2. FlagOS Community
Log inSign up
FlagOS Community
429 posts
Image
user avatar
FlagOS Community
@FlagOS_Official
An open-source system software stack for AI. Bridging Model–System–Chip layers. Build once, run across diverse hardware.
beijing, China
flagos.io/Home?lang=en
Joined December 2024
383
Following
659
Followers
RepliesRepliesMediaMedia
  • Pinned
    user avatar
    FlagOS Community
    @FlagOS_Official
    Jul 29
    At #WAIC2026, FlagOS advanced open compute for AI—from open infrastructure and cross-chip collaboration to the 72-hour operator challenge. Featuring Michael Berns (@AIThoughtLeader), Vincent Caldeira, Mehdi Snene & Lin Yonghua. #FlagOS #OpenSourceAI #AIInfra #OpenComputing
    Image
    00:00
    Image
    Image
    Image
    2.1K
  • user avatar
    FlagOS Community
    @FlagOS_Official
    11h
    What would AI look like if compute had no borders? At #WAIC2026 in Shanghai, we asked global AI and open-source leaders what it will take to build an AI future that is more open, portable and accessible. Their answers point to a shared direction: → Open software stacks that
    Image
    00:00
    108
  • user avatar
    FlagOS Community
    @FlagOS_Official
    Jul 29
    Great to see @Kimi_Moonshot open-source FlashKDA. Building on its two-stage Chunk KDA design, we used Triton-TLE to cut sync idle time and keep the FP32 master state in registers—delivering a 1.38× geomean speedup across 12 H800 workloads.
    user avatar
    Kimi.ai
    @Kimi_Moonshot
    Jul 27
    We've open-sourced FlashKDA, our high-performance CUTLASS-based implementation of Kimi Delta Attention kernels. It delivers 1.72×–2.22× prefill speedup over the flash-linear-attention baseline on H20, and works as a drop-in backend for flash-linear-attention. Explore on GitHub:
    91
  • user avatar
    FlagOS Community
    @FlagOS_Official
    Jul 29
    Long-context speedups aren’t just about complexity—they’re about scheduling and data residency. Building on FlashKDA from @Kimi_Moonshot, our Triton-TLE optimization on @NVIDIAAIDev H800 delivers a 1.38× geomean speedup across 12 workloads. @OpenAIDevs #Triton #AIInfra
    Image
    Image
    Image
    Image
    Made with AI
    1.1K
  • user avatar
    FlagOS Community
    @FlagOS_Official
    Jul 22
    At WAIC 2026, global AI and open-source leaders explored how open compute can break hardware silos, unlock heterogeneous AI infrastructure, and expand access to education, research and innovation worldwide. Read the full recap. #WAIC2026 #FlagOS #OpenSourceAI
    Image
    Image
    Image
    Image
    698
  • See @FlagOS_Official's full profile

    Sign up
    Log in

Log in or sign up for X

See what’s happening and join the conversation

Continue with phone
or
Log in with username or email
Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
Advertisement
Advertisement