TTS-Slots (Piper CPU, Qwen3-TTS auf Ada), M4000 außer Betrieb, Engine-Grenzen im Placement (ADR-0006)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-12 02:12:30 +02:00
parent 0533f224f1
commit f90b7df31c
25 changed files with 586 additions and 23 deletions
@@ -58,10 +58,25 @@ model_list:
      api_key: none
    model_info: {mode: rerank}

+  # --- Sprachausgabe (OpenAI-kompatibles /v1/audio/speech) ---
+  - model_name: tts-de                              # Piper im Infra-LXC (knaster tts-serve)
+    litellm_params:
+      model: openai/tts-de
+      api_base: http://tts-piper:8090/v1
+      api_key: none
+    model_info: {mode: audio_speech}
+  - model_name: tts-de-hq                           # Qwen3-TTS auf gpu-flex (Server-Wahl beim Deployment)
+    litellm_params:
+      model: openai/tts-de-hq
+      api_base: http://gpu-flex:8091/v1
+      api_key: none
+    model_info: {mode: audio_speech}
+
 router_settings:
  # Infrastruktur-Fallbacks innerhalb einer Stufe; Qualitäts-Eskalation macht der Router-Service
  fallbacks:
    - tier1-general: [tier1-flex]
+    - tts-de-hq: [tts-de]
  allowed_fails: 2
  cooldown_time: 60
  num_retries: 1