OpenSpeakers/docker-compose.offline.yml at main · attevon-llc/OpenSpeakers · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
# docker-compose.offline.yml
# Offline / air-gapped deployment overlay.
# Replaces docker-compose.override.yml — no internet, no source mounts, no builds.
#
# Usage:
#   docker compose -f docker-compose.yml -f docker-compose.offline.yml -f docker-compose.gpu.yml up -d
#
# All paths below are relative to the directory containing this file (the install root).
# Expected layout:
#   ./images/               pre-loaded Docker image tarballs (loaded by install.sh)
#   ./model_cache/          HuggingFace + torch cache (populated by download-models.sh)
#   ./vibevoice-repo/       VibeVoice Python package source (for main worker)
#   ./audio_output/         generated audio (created on first run)
#   ./.env                  configured from .env.example

services:
  backend:
    image: open_speakers-backend:latest
    command: uvicorn app.main:app --host 0.0.0.0 --port 8080 --workers 2

  worker:
    image: open_speakers-worker:latest
    volumes:
      - ${MODEL_CACHE_DIR:-./model_cache}/huggingface:/home/appuser/.cache/huggingface
      - ${MODEL_CACHE_DIR:-./model_cache}/torch:/home/appuser/.cache/torch
      - ${AUDIO_OUTPUT_DIR:-./audio_output}:/app/audio_output
      - ${VIBEVOICE_REPO_PATH:-./vibevoice-repo}:/opt/vibevoice
    environment:
      - HF_HOME=/home/appuser/.cache/huggingface
      - HOME=/home/appuser
      - HF_HUB_OFFLINE=1
      - TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_cache
      - USER=appuser
      - PYTHONPATH=/opt/vibevoice

  worker-kokoro:
    image: open_speakers-worker:latest
    volumes:
      - ${MODEL_CACHE_DIR:-./model_cache}/huggingface:/home/appuser/.cache/huggingface
      - ${MODEL_CACHE_DIR:-./model_cache}/torch:/home/appuser/.cache/torch
      - ${AUDIO_OUTPUT_DIR:-./audio_output}:/app/audio_output
    environment:
      - HF_HOME=/home/appuser/.cache/huggingface
      - HOME=/home/appuser
      - HF_HUB_OFFLINE=1
      - TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_cache
      - USER=appuser
      - PYTHONPATH=/app
    command: >
      celery -A app.core.celery worker
      --loglevel=info -Q tts.kokoro
      --pool=threads --concurrency=1
      --max-tasks-per-child=1000
      --hostname kokoro-worker@%h -E

  worker-fish:
    image: open_speakers-worker-fish:latest
    volumes:
      - ${MODEL_CACHE_DIR:-./model_cache}/huggingface:/root/.cache/huggingface
      - ${AUDIO_OUTPUT_DIR:-./audio_output}:/app/audio_output
    environment:
      - PYTHONPATH=/app
      - HF_HOME=/root/.cache/huggingface
      - HF_HUB_OFFLINE=1
      - TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_cache
      - USER=appuser
      - PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
    command: >
      celery -A app.core.celery worker
      --loglevel=info -Q tts.fish-speech
      --pool=threads --concurrency=1
      --hostname fish-worker@%h -E

  worker-qwen3:
    image: open_speakers-worker-qwen3:latest
    volumes:
      - ${MODEL_CACHE_DIR:-./model_cache}/huggingface:/root/.cache/huggingface
      - ${AUDIO_OUTPUT_DIR:-./audio_output}:/app/audio_output
    environment:
      - PYTHONPATH=/app
      - HF_HOME=/root/.cache/huggingface
      - HF_HUB_OFFLINE=1
      - TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_cache
      - USER=appuser
    command: >
      celery -A app.core.celery worker
      --loglevel=info -Q tts.qwen3
      --pool=threads --concurrency=1
      --hostname qwen3-worker@%h -E

  worker-f5:
    image: open_speakers-worker-f5:latest
    volumes:
      - ${MODEL_CACHE_DIR:-./model_cache}/huggingface:/root/.cache/huggingface
      - ${AUDIO_OUTPUT_DIR:-./audio_output}:/app/audio_output
    environment:
      - PYTHONPATH=/app:/opt/cosyvoice:/opt/cosyvoice/third_party/Matcha-TTS
      - HF_HOME=/root/.cache/huggingface
      - HF_HUB_OFFLINE=1
      - TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_cache
      - USER=appuser
    command: >
      celery -A app.core.celery worker
      --loglevel=info -Q tts.f5-tts
      --pool=threads --concurrency=1
      --hostname f5-worker@%h -E

  worker-orpheus:
    image: open_speakers-worker-orpheus:latest
    volumes:
      - ${MODEL_CACHE_DIR:-./model_cache}/huggingface:/root/.cache/huggingface
      - ${AUDIO_OUTPUT_DIR:-./audio_output}:/app/audio_output
    environment:
      - PYTHONPATH=/app
      - HF_HOME=/root/.cache/huggingface
      - HF_HUB_OFFLINE=1
      - TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_cache
      - USER=appuser
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    command: >
      celery -A app.core.celery worker
      --loglevel=info -Q tts.orpheus
      --pool=threads --concurrency=1
      --hostname orpheus-worker@%h -E

  worker-dia:
    image: open_speakers-worker-dia:latest
    volumes:
      - ${MODEL_CACHE_DIR:-./model_cache}/huggingface:/root/.cache/huggingface
      - ${MODEL_CACHE_DIR:-./model_cache}/dac:/root/.cache/descript/dac:ro
      - ${AUDIO_OUTPUT_DIR:-./audio_output}:/app/audio_output
    environment:
      - PYTHONPATH=/app
      - HF_HOME=/root/.cache/huggingface
      - HF_HUB_OFFLINE=1
      - TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_cache
      - USER=appuser
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    command: >
      celery -A app.core.celery worker
      --loglevel=info -Q tts.dia
      --pool=threads --concurrency=1
      --hostname dia-worker@%h -E

  frontend:
    image: open_speakers-frontend:latest
    ports:
      - "${FRONTEND_PORT:-5200}:80"
    healthcheck:
      test: ["CMD", "wget", "--spider", "-q", "http://127.0.0.1:80"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 10s