ryzen-ai

Local inference that behaves like a frontier API: Qwen3.8-Flash-Next on AMD Strix Halo plus small NPU models, over the OpenAI and Anthropic APIs on one port.

The Open Source Strix Halo AI engine. Qwen 3.8 Flash Next Q4_K_XL: 1,801.01 pp, 73.17 tg single user, 194.37 tok/s 8 users; Qwen 3.8 27B Q4_K_XL: 656.33pp, 70.56tg tok/s single user with DFlash2

A close-to-metal Python API for programming AMD Ryzen™ AI NPUs (AI Engines), built on an open-source MLIR-based compiler toolchain.

The fastest way to run Qwen3.8 27B on Strix Halo (gfx1151)

Active and open development fork of FastFlowLM.

Open-source bring-up + verified matmul on the first-gen AMD XDNA1 (Phoenix/Hawk Point) NPU on Linux — the gen FastFlowLM/Lemonade skip. RyzenAI-npu1, mlir-aie/IRON, XRT.

A two-day, non-expert guide to running local LLMs on an AMD Strix Halo (Ryzen AI Max+ 395) box — full ~120GB memory pool, ROCm backend, NPU in parallel via FastFlowLM, one OpenAI-compatible endpoint.

Run LLMs and Ollama Qwen across AMD XDNA1 NPU, CPU, and GPU on Ryzen AI Linux systems.

vLLM + Qwen3.6-27B (BF16) OpenAI-compatible inference server on AMD Strix Halo (Ryzen AI Max+ 395, gfx1151). Vision input, 256K context, /v1/responses with separated reasoning, via TheRock ROCm.

Fedora COPR packaging for FastFlowLM - Run LLMs on AMD Ryzen AI NPUs

Practical applications for the AMD XDNA NPU on Linux — webcam effects (working) + embeddings/audio/ggml roadmap. Built on open-xdna.

Xian-VL monorepo - Multilingual Assistant for Gaming Environments 🧙‍♂️ and other AI translation/research tools. Powered by Lemonade. 🍋

Fix internal speakers on the ASUS ProArt PX13 (HN7306, Strix Halo) under Linux: TAS2783 firmware + ALSA UCM Speaker profile + ACP/SoundWire reset for boot and suspend/resume.

ROCmFPX llama.cpp fork for Windows 🏆 — native build, headless OpenAI-compatible server & benchmarks. Tested on AMD Strix Halo (gfx1151), runs on other GPUs too.

AMD ISP4 camera driver module for Ryzen AI laptops

Stable Diffusion image generation on AMD Ryzen AI NPUs for Linux

llama.cpp + Qwen3.6-27B (Q8_0 GGUF) OpenAI-compatible inference server on AMD Strix Halo (Ryzen AI Max+ 395, gfx1151). 256K context, ~7.5 t/s decode via TheRock ROCm Docker.

Run large LLMs locally on AMD Ryzen AI Max+ 395 (Strix Halo, gfx1151) with ROCmFP4 4-bit quantization. Measured benchmarks, build + serving recipes, and 118 ready-to-run GGUF models.

1bit engine: local LLM inference on the AMD Ryzen AI MAX+ 395 (Strix Halo, gfx1151) inside Lemonade Server as the onebit backend, pinned to Lemonade's weekly release. Radeon 8060S iGPU via HRX with our own HIP kernels, XDNA 2 NPU via our own NPU runtime; OpenAI-compatible API. Qwen3.8, ZAYA1, Ternary-Bonsai, BitNet b1.58, gpt-oss-20b. Apache-2.0.

Zest: speed up long prompts in Lemonade with your AMD Ryzen AI NPU. Works with many GGUF models; direct llama.cpp also supported.