在 AMD Strix Halo 上本地跑 Qwen3.8-Flash-Next 和 NPU 小模型，用同一端口同时提供 OpenAI 与 Anthropic 兼容 API，让本地推理像调前沿云服务一样顺手。
基于 MLIR 的开源编译器工具链，提供贴近硬件的 Python API，用来给 AMD Ryzen AI NPU（AI Engine）编写和编译算子程序。
在本地 GPU 和 NPU 上直接跑优化后的大模型，帮用户发现并启动本地 AI 应用，省去自己折腾推理环境的麻烦。
专为 AMD Ryzen AI NPU 深度优化的本地大模型推理引擎，几分钟内就能在 AMD NPU 上跑起 LLM，省去繁琐配置。
…oved Windows NPU CI and added caching for Windows Ryzen AI wheel builds → @jgmelber - Continued JIT cache correctness, kernel-source handling, documentation, CI, and buil
… wall time, so the 6× reported under Hyper-V on a Ryzen was not reproduced on KVM on an Intel host. The thresholds are removed because they are measured, pure overhead on
…single-thread --workers 1 path — a 3.0x gain (AMD Ryzen AI 9 465, 20 threads, 2-byte body, median of 5 three-second runs). --workers 16 measures slower (280,079 req/s) th
…breHardwareMonitor/LibreHardwareMonitor/pull/1625 Ryzen 9000 Core Frequency Bugfix by @sebastian-dev in https://github.com/LibreHardwareMonitor/LibreHardwareMonitor/pull/
…ormance improved by ≈25% on test PC [AMD RX6600 + Ryzen 5600 + 8 threads] ⊡ On more powerful setups, performance gains may reach up to ≈100% ⊡ Faster and more lightweight