Strata-AMD ghcr.io/niko1221/strata:latest-rocm-gfx120x https://github.com/Niko1221/Strata/pkgs/container/strata latest-rocm-gfx120x RX 9000 series and Radeon AI PRO R9700 (RDNA 4) latest-rocm-gfx110x RX 7000 series (RDNA 3) latest-rocm-gfx1151 Ryzen AI Max "Strix Halo" (Radeon 8060S) 0.2.7-rocm-gfx120x Pinned release (RX 9000 / R9700) 0.2.9-rocm-gfx120x Pinned release (RX 9000 / R9700) bridge bash false https://github.com/Aurial/Unraid-Community-Apps/issues https://github.com/Niko1221/Strata https://raw.githubusercontent.com/Aurial/Unraid-Community-Apps/main/README.md Strata AMD runs the Qwen3.8-Flash-Next mixture-of-experts model on an AMD GPU plus system RAM, and serves an OpenAI- and Anthropic-compatible API, a chat page and a live monitor. This template runs Strata's official image, ROCm build. For NVIDIA cards use the separate Strata template instead. Before you press Apply: by default Strata also copies the model's experts into system RAM, even ones the GPU already holds. On a server that runs other containers this can use most of your memory (measured: about 59 GiB for the container, leaving 4 GiB for everything else). Set Engine settings to resident-budget-gib=8 (two 32 GB cards) and Parallel requests to 2 unless this server is dedicated to Strata. Single-card servers: see the README before choosing a budget. Requirements: the amdgpu kernel driver, which Unraid ships as standard (Linux 6.8+) — no driver plugin needed. Pick your card family under Branch when installing: latest-rocm-gfx120x for the RX 9000 series and Radeon AI PRO R9700 (default), latest-rocm-gfx110x for the RX 7000 series, latest-rocm-gfx1151 for Ryzen AI Max (Strix Halo). The RX 6000 series has no image yet. You also need RAM for the model's experts plus about 10 GB: roughly 44 GB (Q2_0), 46 GB (IQ2_XS), 53 GB (IQ3_XXS) or 60 GB (IQ3_S) total, and room for the model on an SSD pool (about 80 GB IQ3_XXS, 90 GB IQ3_S). First start: downloads the model into the Data folder and prepares it, which takes a while; watch the log. A stopped download resumes where it left off. Later starts go straight to serving. The model name the API serves is listed at http://[IP]:[PORT:8080]/v1/models. RAM: by default Strata also copies the model's experts into system RAM, even ones the cards already hold — a two-card server used about 59 GiB this way. On a busy server cap it under Engine settings: resident-budget-gib=8 keeps in RAM only what the cards do not hold (about 23 GiB measured, with Parallel requests 2, costing about 9% on cold prompt speed). The model fills the GPU memory too: stop other GPU workloads before starting. API key: always on. Leave it empty and the first start makes one and keeps it in the Data folder; show it with: docker exec Strata-AMD strata launch --show-key --preset docker. Clients send it as Authorization: Bearer KEY. Changing settings: the default image tags (latest-rocm-*) follow the maintainer; once your setup is tuned you can pin a version by editing the Repository tag, for example ghcr.io/niko1221/strata:0.2.7-rocm-gfx120x. Edit settings and Apply. The new values replace the saved ones on the next start; the model files and tuning stay in the Data folder. Template and Unraid problems: the Support link. Strata itself: the Project link. AI: llm qwen local ai openai anthropic api inference chat amd rocm radeon rdna gpu http://[IP]:[PORT:8080]/ https://raw.githubusercontent.com/Aurial/Unraid-Community-Apps/main/templates/strata-amd.xml https://raw.githubusercontent.com/Aurial/Unraid-Community-Apps/main/icon.png MIT --device /dev/kfd --device /dev/dri --group-add video --ulimit memlock=-1:-1 --shm-size=2g --stop-timeout 120 **amdgpu** kernel driver — Unraid ships it, no driver plugin needed (Linux 6.8+). An RDNA 3 / RDNA 4 / Strix Halo card: RX 7000 series, RX 9000 series, Radeon AI PRO R9700, or Ryzen AI Max — pick the matching **Branch** at install (gfx120x / gfx110x / gfx1151); the RX 6000 series has no image yet. Do not set a memory limit on the container: Strata plans from the server's whole RAM. Do not pin CPUs: the engine sets its own thread affinity. 8080 /mnt/cache/appdata/strata-amd all 128k off skip 0 int8