nemotron-unified-wyoming ghcr.io/marco-taylor/nemotron-unified-wyoming:latest https://ghcr.io/marco-taylor/nemotron-unified-wyoming bridge false https://github.com/marco-taylor/nemotron-unified-wyoming/issues https://github.com/marco-taylor/nemotron-unified-wyoming https://raw.githubusercontent.com/marco-taylor/nemotron-unified-wyoming/main/config/unraid-template.xml Deutsche CPU-Spracherkennung mit NVIDIA Nemotron Labs Unified ASR 0.6B L37 INT8 über Wyoming. Fehlende oder beschädigte Modelle werden automatisch aus dem festen GitHub-Modell-Release models-l37-v1.0.0 geladen und mit SHA256 geprüft; die Erstinitialisierung kann länger dauern (ca. 693 MiB Modelldownload). Gültige vorhandene Modelle werden ohne erneuten Download verwendet. Persistenter Modellcache, PUID/PGID und eingeschränkter STT-Prozess. Modelllizenz OpenMDW-1.1; NVIDIA trainierte das Modell, dieses Projekt übernimmt ONNX-Export und partielle Encoderquantisierung. Kein offizieller Server-v1.0.0-Release. HomeAutomation: https://raw.githubusercontent.com/marco-taylor/nemotron-unified-wyoming/main/assets/icons/nemotron-lowq.png --read-only --cap-drop=ALL --cap-add=SETUID --cap-add=SETGID --security-opt=no-new-privileges:true --memory=4g --cpus=4 --pids-limit=128 --stop-timeout=60 --tmpfs /tmp:size=16m,noexec,nosuid,nodev --log-opt max-size=10m --log-opt max-file=3 serve 10310 /mnt/user/appdata/nemotron-unified-wyoming/models/L37-runtime 2 120 8 30 info false 99 100