Join FlagOS Community on Slack — 开发者交流频道现已开放

[wbavon]

We have opened FlagOS Community on Slack for day-to-day developer conversations. Start in #community; #kernels-and-compilers and #models-and-serving are optional technical channels.

Join FlagOS Community

If you work on Triton kernels, GPU/AI accelerator enablement, vLLM/SGLang/PyTorch integration, distributed communication, training or inference performance, you are welcome to join. English and Chinese are both welcome.

Introduce yourself with your project or GitHub profile, the hardware/software you work with, and one question or idea you would like to discuss. You do not need to be a member of the flagos-ai GitHub organization. There is no administrator approval step; Slack's normal account sign-in/email verification still applies.

Please use the invitation link above when joining for the first time. flagos-ai.slack.com is the workspace sign-in address for existing members. If a mobile app opens the ordinary sign-in screen, open the full invitation link in a browser instead.

Useful starting points: FlagGems, FlagTree, FlagScale, FlagCX, and our contribution guide.

Please keep durable bug reports, design proposals and code reviews on GitHub, and follow our Code of Conduct.

---

众智 FlagOS Community 的 Slack 已开放，从 #community 开始交流，也可按兴趣加入 #kernels-and-compilers 或 #models-and-serving。欢迎做 Triton/GPU 算子、AI 芯片适配、vLLM/SGLang/PyTorch 集成、通信和训推性能优化的开发者加入，中英文均可。

点击邀请链接加入。无需 GitHub 组织权限、无需管理员审批；需要完成 Slack 正常登录或邮箱验证。欢迎加入后介绍自己的项目、硬件和正在解决的问题，也可以把邀请转给方向相关的开发者。

Public technical resources

Reading Triton benchmarks: kernel time, full API cost, and inference performance includes a public case study and a compact template for comparing measurements. Read it here on GitHub, and join the conversation wherever it works for you.

Same shape, different layout provides four input layouts to preserve in an operator-correctness reproducer. Both notes are public guides, not new device benchmark results.

Three concrete ways to participate

Checked on 22 October 2026, Beijing time. Read the latest issue and related PR before starting; these are investigation and review opportunities, not assigned tasks or labeled beginner issues.

• Operator correctness: FlagGems #7396 has a zero-stride gradient reproducer and contiguous control. Review the reference expectations without a GPU, or run it on a supported device and report exact versions, strides and results. The reporter's RTX 2080 Ti results do not establish behavior on every current backend.
• Packaging and compiler integration: FlagTree #1392 already has open PR #1393. Help review its package-discovery cases or test assumptions instead of duplicating the patch. Source/import checks can start without a GPU; working device communication needs runtime and hardware validation.
• Training setup and documentation: FlagScale #1295 reports a tokenizer-path assertion. Compare the image tag, installed versions and selected YAML with the current getting-started guide, which already documents tokenizer_path. This configuration audit can start without a GPU; it is not evidence that the latest build still fails.

Keep findings and proposed fixes on the original GitHub threads. Slack is optional for comparing environments and approaches; there are no bounties or promised fixes attached to these suggestions.

[wbavon]

The Slack entry for the community README is also prepared in #154 (English and Chinese). As requested by the bootstrap-phase guidance in MAINTAINERS.md, I’m posting here to request a community maintainer review. The organization homepage already has the live invitation; the README PR remains open for review.