Repositories (11)
llama.cpp OpenAI-compatible server on Vulkan for AMD Strix Halo (gfx1151), GGUF weights pinned to GTT not VRAM. Serves poolside Laguna S 2.1, Gemma 4 and Qwen3.6 GGUFs on the stock Vulkan image, plus an opt-in ROCmFP4 + MTP stack (Ubuntu 26.04 + TheRock ROCm 7.13). Docker Compose, with real measured benchmarks.
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.
Self-hosted, browser-based AI dungeon RPG, fully local on an AMD Strix Halo APU: narrator + per-character agents on one local LLM (Gemma 4 26B MoE via llama.cpp/Vulkan), local images (FLUX.2 klein/ComfyUI), expressive voice (Maya1 TTS). FastAPI brain, SQLite, vanilla JS, docker-compose with guided setup (CLI wizard or double-click HTML).
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.
ComfyUI on AMD Strix Halo (RDNA 3.5 / gfx1151) via Docker. Ubuntu 26.04 LTS + uv-managed Python 3.12 + pinned TheRock ROCm 7.13 wheels. Fixes the silent CPU fallback Debian / Python 3.13 images hit on gfx1151.
ROCm-first OpenAI-compatible inference gateway for open-weight reasoning models. Single active profile (gpt-oss-20b, deepseek-r1-distill, qwen3-4b), host-mounted weights, fail-closed contract. FastAPI + vLLM.
Local AI album generation: LLM album plan + ACE-Step 1.5 track generation + FLUX cover art, all on AMD ROCm. SSE-streamed FastAPI backend, React frontend, MP3/MP4 + YouTube-ready exports. Docker Compose.
Docker infrastructure for AMD Strix Halo (RDNA 3.5 / gfx1151): PyTorch + ROCm base container and a separate Ollama LLM service. Two folders, two Compose files, one Strix Halo box.
Text to a rigged, low-poly GLB as a local Claude skill on an AMD Strix Halo APU: FLUX.2 klein via ComfyUI, TRELLIS.2 on a Vulkan-only engine, Blender headless for a measured skeleton with idle/walk/run/jump plus a pose check. Ships a three.js gallery that renders every asset, CC0 search and an MCP server. 226 tests.
Agentic control layer for the Censurado self-hosted AI news portal: a stdlib Python CLI plus an MCP server (32 tools) an LLM agent drives to write and publish AI-persona articles through a gated editorial walk, curate the front page and authors, and deploy. Ships the prompt/skill recipe and a 24/7 serve loop.
Production-oriented Docker Compose stack serving openai/gpt-oss-20b via vLLM on AMD Strix Halo (gfx1151, ROCm 7.2). OpenAI Responses API, host-mounted weights, hard-capped KV cache. Verified, no source build.