Remove Beta 1 and Piper fallback

compose.yaml

@@ -230,89 +230,6 @@ services:
       - --spec-draft-p-min
       - "0.05"
 
-  # Beta 1 keeps the complete GSQ-RCO IQ3_S text model and KV cache on the RTX
-  # 5080. Only the multimodal projector runs on the RTX 3060. The larger IQ3_S
-  # build starts conservatively at 112K until its hard context boundary has
-  # been measured; the former IQ3_XXS result does not transfer to this file.
-  llama-beta1:
-    <<: *llama-common
-    container_name: mike-ai-llama-beta1
-    labels:
-      com.mike-ai.llama-profile: beta1
-    environment:
-      NVIDIA_VISIBLE_DEVICES: ${BETA1_GPU_DEVICES:-0,1}
-      NVIDIA_DRIVER_CAPABILITIES: compute,utility
-      MTMD_BACKEND_DEVICE: CUDA1
-    command:
-      - --model
-      - "/models/${BETA1_MODEL_FILE:?BETA1_MODEL_FILE is required}"
-      - --mmproj
-      - "/models/${VISION_PROJECTOR_FILE:?VISION_PROJECTOR_FILE is required}"
-      - --mmproj-offload
-      - --mmproj-device
-      - CUDA1
-      - --alias
-      - qwen-beta-1
-      - --ctx-size
-      - "${BETA1_CONTEXT:-112000}"
-      - --flash-attn
-      - "on"
-      - --cache-type-k
-      - q4_0
-      - --cache-type-v
-      - q4_0
-      - --cache-prompt
-      - --cache-ram
-      - "${LLAMA_CACHE_RAM_MIB:-32768}"
-      - --threads
-      - "${LLAMA_THREADS:-6}"
-      - --threads-batch
-      - "${LLAMA_THREADS_BATCH:-6}"
-      - --batch-size
-      - "${BETA1_BATCH_SIZE:-2048}"
-      - --ubatch-size
-      - "${BETA1_UBATCH_SIZE:-128}"
-      - --parallel
-      - "${BETA1_PARALLEL_SLOTS:-1}"
-      - --kv-unified
-      - --jinja
-      - --reasoning
-      - auto
-      - --reasoning-preserve
-      - --host
-      - 0.0.0.0
-      - --port
-      - "8080"
-      - --metrics
-      - --fit
-      - "off"
-      - --n-gpu-layers
-      - all
-      - --no-mmap
-      - --no-ui
-      - --temperature
-      - "1.0"
-      - --top-p
-      - "0.95"
-      - --top-k
-      - "20"
-      - --device
-      - CUDA0
-      - --main-gpu
-      - "0"
-      - --split-mode
-      - none
-      - --spec-type
-      - draft-mtp
-      - --spec-draft-n-max
-      - "3"
-      - --spec-draft-type-k
-      - f16
-      - --spec-draft-type-v
-      - f16
-      - --spec-draft-p-min
-      - "0.05"
-
   llama-large:
     <<: *llama-common
     container_name: mike-ai-llama-large
@@ -565,7 +482,7 @@ services:
       - /var/run/docker.sock:/var/run/docker.sock
     environment:
       CONTROLLER_TOKEN: "${CONTROLLER_TOKEN:?CONTROLLER_TOKEN is required}"
-      ALLOWED_PROFILES: fast,medium,beta1,large,ultra,uncensored
+      ALLOWED_PROFILES: fast,medium,large,ultra,uncensored
       IMAGE_WORKER: image
       RESTORE_WORKER: restore
       TTS_WORKER: qwen3
@@ -622,11 +539,10 @@ services:
       CHAT_IMAGE_ALLOW_REMOTE_URLS: "false"
       ENABLE_IMAGE_GENERATION: "true"
       ENABLE_TTS: "true"
-      # Stable OpenAI compatibility names remain piper/alloy for existing
-      # clients. The gateway maps them to Qwen3-TTS and falls back to Piper
-      # while Qwen3-TTS is unavailable, busy or paused for image generation.
+      # The gateway keeps text normalization, output conversion and native
+      # PCM streaming in one stable API in front of Qwen3-TTS.
       TTS_WORKER_URL: http://tts-gateway:8085
-      TTS_MODEL: piper
+      TTS_MODEL: qwen3-tts
       TTS_VOICES: alloy
       TTS_DEFAULT_VOICE: alloy
       ENABLE_STT: "true"
@@ -655,8 +571,6 @@ services:
         condition: service_healthy
       profile-controller:
         condition: service_healthy
-      piper:
-        condition: service_healthy
       tts-gateway:
         condition: service_healthy
       whisper:
@@ -699,37 +613,6 @@ services:
       timeout: 3s
       retries: 12
 
-  piper:
-    build:
-      context: platform/docker/piper
-      args:
-        PIPER_TTS_VERSION: ${PIPER_TTS_VERSION:-1.6.0}
-    image: mike-ai/piper:local
-    container_name: mike-ai-piper
-    restart: unless-stopped
-    read_only: true
-    tmpfs:
-      - /tmp:size=256m,mode=1777
-    volumes:
-      - piper-data:/data
-    environment:
-      PIPER_DATA_DIR: /data
-      PIPER_VOICE: ${PIPER_VOICE:-de_DE-thorsten-high}
-      PIPER_VOICE_ALIAS: alloy
-      PIPER_HOST: 0.0.0.0
-      PIPER_PORT: "8085"
-      PIPER_MAX_TEXT_CHARS: "8000"
-    networks: [frontend]
-    security_opt: ["no-new-privileges:true"]
-    cap_drop: [ALL]
-    cap_add: [CHOWN, SETUID, SETGID]
-    healthcheck:
-      test: [CMD, curl, -fsS, "http://127.0.0.1:8085/status"]
-      interval: 10s
-      timeout: 5s
-      retries: 30
-      start_period: 120s
-
   qwen3-tts:
     image: ${QWEN3_TTS_IMAGE:-ghcr.io/malaiwah/qwen3-tts-server:latest@sha256:b363a01d08b1bbecbfc3ca6f585368fae2cfdc591f9ecca6643738369f9a9d98}
     container_name: mike-ai-qwen3-tts
@@ -787,17 +670,12 @@ services:
       QWEN_TTS_VOICE: serena
       QWEN_TTS_LANGUAGE: German
       QWEN_TTS_TIMEOUT: "120"
-      PIPER_URL: http://piper:8085
       TTS_VOICE_ALIAS: alloy
       TTS_DEFAULT_LANGUAGE: de
       # Mixed-language clip stitching caused long pauses and unintelligible
       # transitions. Keep full sentences in one stable German voice.
       TTS_CODE_SWITCH_ENABLED: "false"
-      PIPER_TIMEOUT: "120"
     networks: [frontend]
-    depends_on:
-      piper:
-        condition: service_healthy
     security_opt: ["no-new-privileges:true"]
     cap_drop: [ALL]
     healthcheck:
@@ -920,7 +798,6 @@ services:
       - /data/docker-backups:/archive
       - /etc/mike-ai:/backup/etc-mike-ai:ro
       - /opt/mike-ai/stack:/backup/stack:ro
-      - piper-data:/backup/volumes/piper-data:ro
       - router-state:/backup/volumes/router-state:ro
       - router-images:/backup/volumes/router-images:ro
       - portainer-data:/backup/volumes/portainer-data:ro
@@ -947,7 +824,6 @@ networks:
     name: mike-ai-tools-egress
 
 volumes:
-  piper-data:
   whisper-data:
   router-state:
   router-images: