FlagOS Community

Open-source AI system software: build once, run across diverse hardware.

[FlagOS Community @FlagOSCommunity]

🎁 20 WINNERS. 2 PRIZE POOLS.

⚡ 1000+ AI Compute Credits — explore ChatGPT, Claude Code, DeepSeek, GLM & more
🎟️ 10× Open Computing Conference tickets · Beijing · Oct 17–18

How to enter:
✅ Follow @FlagOSCommunity
That’s it. You’re in.

Ends Sep 24, 23:59 UTC. #FlagOS #AI

[FlagOS Community @FlagOSCommunity]

Winners will be randomly selected from followers after the giveaway ends and announced via @FlagOSCommunity. Let’s get started! ✅

[FlagOS Community @FlagOSCommunity]

🎁 20 WINNERS. 2 PRIZE POOLS.

⚡ 1000+ AI Compute Credits — explore ChatGPT, Claude Code, DeepSeek, GLM & more
🎟️ 10× Open Computing Conference tickets · Beijing · Oct 17–18

How to enter:
✅ Follow @FlagOSCommunity
That’s it. You’re in.

Ends Sep 24, 23:59 UTC. #FlagOS #AI

[FlagOS Community @FlagOSCommunity]

🚀 FlagOS × MiniCPM5-2B Inference Throughput Challenge is LIVE!

Push total tokens/s on Iluvatar CoreX BI-V150 + MetaX XiYun C500.

🏆 Up to ¥30K
📅 Register by Nov 10
🔗 https://t.co/NdHWE60oLU

#FlagOS #MiniCPM #LLMInference @IEEEorg @OpenBMB

[FlagOS Community @FlagOSCommunity]

20+ hardware backends. 800+ open-source ops. Day 0 multi-chip adaptation.
At our Shanghai forum alongside #KubeCon + @PyTorch Conference China, FlagOS showed how open collaboration can break AI hardware fragmentation.

@linuxfoundation @CloudNativeFdn #OpenSourceAI #AIInfra

[FlagOS Community @FlagOSCommunity]

Long-context inference slowing your AI down?

Triton-TLE optimizes Kimi’s KDA in two places:
⚡ K1: shared memory + async loading
⚡ K2: register-resident states + WGMMA

Result: 1.38× avg speedup on H800, ~1.4× in key long-context workloads.

🔗 GitHub: https://t.co/otIPKHpYaf

[FlagOS Community @FlagOSCommunity]

🔗Full Article:https://t.co/Lcm96R1uMa

[FlagOS Community @FlagOSCommunity]

One RL codebase. More hardware. ⚡

veRL 0.9.0 × FlagOS brings multi-chip adaptation to RL post-training via verl-hardware-plugin—keeping training logic hardware-agnostic and lowering the cost of adding new accelerators.

#veRL #FlagOS #RLHF #AIInfra @verl_project @BytedanceTalk

[FlagOS Community @FlagOSCommunity]

Calling all @sgl_project builders 👀

Pick a real SGLang kernel. Beat the baseline. Climb the multi-chip leaderboard.

We’re kicking off #MakeSGLangFaster — a global challenge for Triton & inference hackers.

Who’s in? ⚡

[FlagOS Community @FlagOSCommunity]

Can you make SGLang faster? ⚡

200+ real inference kernels. Triton/Triton-TLE. Multi-chip evaluation. Live leaderboard.

Join the FlagOS Open Computing Global Challenge.
🌍 Worldwide | Teams up to 5
📅 Register by Nov 13
🔗 https://t.co/VCU1gsxSfJ

#SGLang #Triton

[FlagOS Community @FlagOSCommunity]

Can you make SGLang faster? ⚡

200+ real inference kernels. Triton/Triton-TLE. Multi-chip evaluation. Live leaderboard.

Join the FlagOS Open Computing Global Challenge.
🌍 Worldwide | Teams up to 5
📅 Register by Nov 13
🔗 https://t.co/VCU1gsxSfJ

#SGLang #Triton

[FlagOS Community @FlagOSCommunity]

At BAAI in Beijing, the UNDP-led timbuktoo China–Africa AI seminar brought together 50+ research & industry representatives to explore locally adapted AI, edge computing and open infrastructure.

@UNDPChina @UNDPAfrica

#AI #ChinaAfrica #OpenSource

[FlagOS Community @FlagOSCommunity]

Open AI computing is moving beyond single-vendor stacks.

On Sept 7 in Shanghai, FlagOS brings open-source, framework & hardware leaders together to discuss open software for diverse AI hardware.

Co-located with @CloudNativeFdn, @openinfradev & @PyTorch.

#OpenSource #AI