[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-94654":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":9,"totalLinesOfCode":9,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":9,"subscribersCount":16,"size":16,"stars1d":16,"stars7d":16,"stars30d":17,"stars90d":16,"forks30d":16,"starsTrendScore":16,"compositeScore":18,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":19,"hasPages":19,"topics":21,"createdAt":9,"pushedAt":9,"updatedAt":27,"readmeContent":28,"aiSummary":29,"trendingCount":16,"starSnapshotCount":16,"syncStatus":30,"lastSyncTime":31,"discoverSource":32},94654,"FastFlowLM","ROCm\u002FFastFlowLM","ROCm","Run LLMs on AMD Ryzen™ AI NPUs in minutes. Just like Ollama - but purpose-built and deeply optimized for the AMD NPUs.",null,"https:\u002F\u002Fgithub.com\u002FROCm\u002FFastFlowLM","C++",1724,137,25,106,0,8,52.22,false,"main",[22,23,24,25,26],"amd","deepseek","llama","llm","npu","2026-08-24 04:01:22","\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fwww.fastflowlm.com\" target=\"_blank\">\n    \u003Cimg src=\"assets\u002Flogo.png\" alt=\"FastFlowLM Logo\" width=\"200\"\u002F>\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNPU-Optimized-red\" \u002F>\n\u003C\u002Fp>\n\n## ⚡ FastFlowLM (FLM) — Unlock Ryzen™ AI NPUs\n\nRun large language models — now with **Vision**, **Audio**, **Embedding** and **MoE** support — on **AMD Ryzen™ AI NPUs** in minutes.  \n**No GPU required. Faster and over 10× more power-efficient. Supports context lengths up to 256k tokens. Ultra-Lightweight (17 MB). Installs within 20 seconds.**\n\n📦 **The only out-of-box, NPU-first runtime built exclusively for Ryzen™ AI.**  \n🤝 **Think Ollama — but deeply optimized for NPUs.**  \n✨ **From Idle Silicon to Instant Power — FastFlowLM Makes Ryzen™ AI Shine.**\n\n> FastFlowLM (FLM) supports all Ryzen™ AI Series chips with XDNA2 NPUs (Strix, Strix Halo, Kraken, and Gorgon Point).\n\n---\n\n## 🔗 Quick Links\n\n  🔽 **[Download](https:\u002F\u002Fgithub.com\u002FROCm\u002FFastFlowLM\u002Freleases\u002Flatest\u002Fdownload\u002Fflm-setup.msi)** | 📊 **[Benchmarks](https:\u002F\u002Ffastflowlm.com\u002Fdocs\u002Fbenchmarks\u002F)** | 📦 **[Model List](https:\u002F\u002Ffastflowlm.com\u002Fdocs\u002Fmodels\u002F)**  \n\n  🐧 **[Linux Getting Started Guide](.\u002Fdocs\u002Flinux-getting-started.md)**\n\n  📖 **[Docs](https:\u002F\u002Ffastflowlm.com\u002Fdocs)** | 📺 **[Demos](https:\u002F\u002Fwww.youtube.com\u002Fplaylist?list=PLf87s9UUZrJoDdz639Yc6w1UTyJ4cFHZ1)** | 💬 **[Discord](https:\u002F\u002Fdiscord.gg\u002Fz24t23HsHF)** \n\n---\n\n## 🚀 Quick Start\n\nA packaged FLM Windows installer is available here: [**flm-setup.msi**](https:\u002F\u002Fgithub.com\u002FROCm\u002FFastFlowLM\u002Freleases\u002Flatest\u002Fdownload\u002Fflm-setup.msi). For more details, see the [release notes](https:\u002F\u002Fgithub.com\u002FROCm\u002FFastFlowLM\u002Freleases\u002F).\n\n📺 [**Watch the quick start video (Windows)**](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=mYOfDNkyBII)\n\n> [!IMPORTANT]  \n> ⚠️ Ensure NPU driver version is **>= 32.0.203.304** (`.304` is the minimum requirement but `.311` is recommended; check via Task Manager→Performance→NPU or Device Manager).  \n> ⚙️ **Tip:**\n>   * **RECOMMENDED**: Try running **Windows Update** or **[Driver Download](https:\u002F\u002Fwww.amd.com\u002Fen\u002Fsupport)**.\n>   * **[Official AMD Install Doc](https:\u002F\u002Fryzenai.docs.amd.com\u002Fen\u002Flatest\u002Finst.html#install-npu-drivers)** *(AMD account required)*.\n>   * **[Unofficial forum downloads](https:\u002F\u002Fwww.elevenforum.com\u002Ft\u002Fdrivers-amd-npu-ryzen-8xxx-9xxx-apu.24220\u002F)** *(CAUTION: third-party content not verified by AMD; download and use at your own risk)*.\n\nAfter installation, open **PowerShell** (`Win + X → I`). To run a model in terminal (**CLI Mode**):\n```powershell\nflm run llama3.2:1b\n```\n> **Notes:**\n> - Internet access to HuggingFace is required to download the optimized model kernels.\n> - Sometimes downloads from HuggingFace may get corrupted. If this happens, run `flm pull \u003Cmodel_tag> --force` (e.g. `flm pull llama3.2:1b --force`) to re-download and fix them.\n> - By default, models are stored in:\n>   - **Windows**: `C:\\Users\\\u003CUSER>\\Documents\\flm\\models\\`\n>   - **Linux**: `~\u002F.config\u002Fflm\u002F`\n> - During installation on Windows, you can select a different base folder (e.g., if you choose `C:\\Users\\\u003CUSER>\\flm`, models will be saved under `C:\\Users\\\u003CUSER>\\flm\\models\\`).\n> - On Linux, you can override the default location by setting the `FLM_MODEL_PATH` environment variable.\n> - To disable the startup version check, set `FLM_DISABLE_UPDATE_CHECK=1`.\n> - ⚠️ If HuggingFace is not accessible in your region, manually download the model ([check this issue](https:\u002F\u002Fgithub.com\u002FROCm\u002FFastFlowLM\u002Fissues\u002F2)) and place it in the chosen directory.   \n\n🎉🚀 FastFlowLM (FLM) is ready — your NPU is unlocked and you can start chatting with models right away!\n\nOpen **Task Manager** (`Ctrl + Shift + Esc`). Go to the **Performance** tab → click **NPU** to monitor usage.  \n\n> **⚡ Quick Tips:**  \n> - Use `\u002Fverbose` during a session to turn on performance reporting (toggle off with `\u002Fverbose` again).   \n> - Type `\u002Fbye` to exit a conversation.  \n> - Run `flm list` in PowerShell to show all available models.  \n\nTo start the local server (**Server Mode**):\n```powershell\nflm serve llama3.2:1b\n```\n> The model tag (e.g., `llama3.2:1b`) sets the initial model, which is optional. If another model is requested, FastFlowLM will automatically switch to it. Local server is on port 52625 (default).  \n\n**[![FastFlowLM Docs](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FFastFlowLM-Detailed%20Instructions-red?style=flat&logo=readthedocs)](https:\u002F\u002Ffastflowlm.com\u002Fdocs\u002Finstructions\u002F)**\n\n---\n\n## 📰 In the News\n\n- 08\u002F11\u002F2026 🎉 FLM is now part of **[ROCm](https:\u002F\u002Fgithub.com\u002FROCm\u002FFastFlowLM)** (v1.0.0) — the repo has moved to AMD's open-source ROCm organization.\n\n- 08\u002F11\u002F2026 🎉 FLM releases its first **SmolVLA** model (v1.0.0) — a Vision-Language-Action robotics policy running on the NPU. See the **[model card](https:\u002F\u002Ffastflowlm.com\u002Fdocs\u002Fmodels\u002Fsmolvla\u002F)** and **[benchmarks](https:\u002F\u002Ffastflowlm.com\u002Fdocs\u002Fbenchmarks\u002Fsmolvla_results\u002F)**.\n\n- 07\u002F17\u002F2026 🎉 FLM is now part of AMD **[news](https:\u002F\u002Fwww.amd.com\u002Fen\u002Fblogs\u002F2026\u002Ffastflowlm-joins-amd-to-advance-ai-inference.html)**. Read **[our story](.\u002Fflm_story.md)** — from a 2025 university project to AMD.\n\n- 03\u002F11\u002F2026 🎉 FLM now supports Linux 🐧 ! To get started, check out the **[quick start guide](https:\u002F\u002Ffastflowlm.com\u002Fdocs\u002Finstall_lin\u002F)** or the **[Lemonade Server docs](https:\u002F\u002Flemonade-server.ai\u002Fflm_npu_linux.html)**, and watch the **[short video](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=tXRchP3sKA8)** for a quick walkthrough of FLM on Linux via Lemonade 🍋.\n\n- 10\u002F01\u002F2025 🎉 FLM was integrated into AMD's **[Lemonade Server](https:\u002F\u002Flemonade-server.ai\u002F)** 🍋. Watch this **[short demo](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=w0Tb3h4WUnE)** about using FLM in Lemonade.\n\n---\n\n## 🧠 Local AI on NPU\n\nFLM makes it easy to run cutting-edge **LLMs** (and now **VLMs**) locally with:\n- ⚡ Fast and low power\n- 🧰 Simple CLI and API (REST and OpenAI API)\n- 🔐 Fully private and offline\n\nNo model rewrites, no tuning — it just works.\n\n---\n\n## ✅ Highlights\n\n- **Runs fully on AMD Ryzen™ AI NPU** — no GPU or CPU load\n- **Lightweight runtime (17 MB)** — installs within **20 seconds**, easy to integrate    \n- **Developer-first flow** — like Ollama, but optimized for NPU  \n- **Support for long context windows** — up to 256k tokens (e.g., Qwen3-4B-Thinking-2507)  \n- **No low-level tuning required** — You focus on your app, we handle the rest\n\n---\n\n## 📄 License\n\n- All orchestration code and CLI tools are open-source under the [MIT License](.\u002FLICENSE_RUNTIME.txt).  \n- These NPU-accelerated binary kernels are completely free for any use, including commercial use.\n- Please acknowledge FastFlowLM in your README\u002Fproject page (or product) as follows:\n  ```\n  Powered by [FastFlowLM](https:\u002F\u002Fgithub.com\u002FROCm\u002FFastFlowLM)\n  ```\n  \n---\n\n💬 Have **feedback\u002Fissues** or want **early access** to our new releases? [Open an issue](https:\u002F\u002Fgithub.com\u002FROCm\u002FFastFlowLM\u002Fissues\u002Fnew) or [Join our Discord community](https:\u002F\u002Fdiscord.gg\u002Fz24t23HsHF)\n\n---\n\n## 🙏 Acknowledgements\n\n- Powered by the advanced **AMD Ryzen™ AI NPU architecture**\n- Inspired by the widely adopted [llama.cpp](https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp) and [Ollama](https:\u002F\u002Fgithub.com\u002Follama\u002Follama)\n- Tokenization accelerated with [MLC-ai\u002Ftokenizers-cpp](https:\u002F\u002Fgithub.com\u002Fmlc-ai\u002Ftokenizers-cpp)\n- Chat formatting via [Google\u002Fminja](https:\u002F\u002Fgithub.com\u002Fgoogle\u002Fminja)\n- Low-level kernels optimized using the powerful [IRON](https:\u002F\u002Fgithub.com\u002Famd\u002Firon)+[AIE-MLIR](https:\u002F\u002Fgithub.com\u002FXilinx\u002Fmlir-aie)\n\n---\n\n## 🛠️ Building from Source\n\nFor developers who want to build FastFlowLM from source, we provide CMake presets for a convenient and consistent build experience.\n\n### Prerequisites\n\n- Git\n- CMake (version 3.22 or higher)\n- A C++20 compatible compiler (e.g., GCC, Clang, MSVC)\n- Ninja (recommended)\n\n### Build Instructions\n\nMore details on the exact procedure, with dependencies to be installed, for linux can be found in [linux-getting-started.md](docs\u002Flinux-getting-started.md).\n\n1.  **Clone the repository:**\n\n    ```bash\n    git clone --recursive https:\u002F\u002Fgithub.com\u002FROCm\u002FFastFlowLM.git\n    cd FastFlowLM\u002Fsrc\n    ```\n\n2.  **Configure CMake using presets:**\n\n    -   **For Linux:**\n\n        ```bash\n        cmake --preset linux-default\n        ```\n\n        This will configure the build to install to `\u002Fopt\u002Ffastflowlm`.\n\n    -   **For Windows (in a developer command prompt):**\n\n        ```bash\n        cmake --preset windows-default\n        ```\n\n3.  **Build the project:**\n\n    ```bash\n    cmake --build build\n    ```\n\n4.  **Install the project (optional):**\n\n    -   **For Linux:**\n\n        ```bash\n        sudo cmake --install build\n        ```\n\n    -   **For Windows (with administrator privileges):**\n\n        ```bash\n        cmake --install build\n        ```\n","FastFlowLM 是一个专为 AMD Ryzen™ AI NPU（XDNA2 架构）优化的轻量级大语言模型（LLM）推理运行时。它支持文本、视觉、音频、嵌入及 MoE 模型，上下文长度最高达 256k tokens，安装包仅 17 MB，可在 20 秒内完成部署；无需独立 GPU，功耗降低超 10 倍。项目基于 C++ 实现，深度集成 ROCm 工具链，提供类 Ollama 的简洁 CLI 接口（如 `flm run llama3.2:1b`），并自动拉取 NPU 专用优化模型。适用于在搭载 Strix\u002FHalo\u002FKraken\u002FGorgon Point 等 Ryzen AI APU 的 Windows\u002FLinux 设备上进行本地、低功耗、实时 LLM 推理。",2,"2026-08-13 02:30:10","trending"]