MiniCPM-o-Demo/docker-compose.multi.yml at main · OpenBMB/MiniCPM-o-Demo · 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
# 一卡多 worker 变体:每张 GPU 放 2 个 worker-backend(共 4 个并发)。
#
# 与主 docker-compose.yml 的唯一区别:GPU0 上放 0a/0b 两个实例、GPU1 上放 1a/1b 两个,
# device_ids 成对指向同一张卡。端口无需错开(各自独立容器,容器内端口默认即可)。
#
# 显存:每实例独立加载一份模型(~21.8GB),每卡 2 份 ≈ 43.6GB,98GB 卡轻松。
# 代价见 README "一张 GPU 跑多个 Worker":N 份重复权重 + 同卡算力瓜分。
#
# 用法:
#   MODEL_HOST_PATH=/path/to/MiniCPM-o-4_5 docker compose -f docker-compose.multi.yml up -d --build

x-worker-backend: &worker-backend-base
  build:
    context: .
    dockerfile: docker/Dockerfile.worker-backend
  image: minicpm-wb:dev
  environment:
    GPU_ID: "0"
    BACKEND_PORT: "22500"
    WORKER_PORT: "22400"
    MODEL_PATH: /models/MiniCPM-o-4_5
  restart: unless-stopped
  healthcheck:
    test: ["CMD", "curl", "-sf", "http://127.0.0.1:22400/health"]
    interval: 15s
    timeout: 5s
    retries: 40
    start_period: 120s
  logging:
    driver: json-file
    options: {max-size: "50m", max-file: "3"}

services:
  worker-backend-0a:
    <<: *worker-backend-base
    container_name: minicpm-wb-0a
    volumes:
      - ${MODEL_HOST_PATH:?set MODEL_HOST_PATH to the host model dir}:/models/MiniCPM-o-4_5:ro
    deploy:
      resources:
        reservations:
          devices:
            - {driver: nvidia, device_ids: ["0"], capabilities: [gpu]}

  worker-backend-0b:
    <<: *worker-backend-base
    container_name: minicpm-wb-0b
    volumes:
      - ${MODEL_HOST_PATH:?set MODEL_HOST_PATH to the host model dir}:/models/MiniCPM-o-4_5:ro
    deploy:
      resources:
        reservations:
          devices:
            - {driver: nvidia, device_ids: ["0"], capabilities: [gpu]}

  worker-backend-1a:
    <<: *worker-backend-base
    container_name: minicpm-wb-1a
    volumes:
      - ${MODEL_HOST_PATH:?set MODEL_HOST_PATH to the host model dir}:/models/MiniCPM-o-4_5:ro
    deploy:
      resources:
        reservations:
          devices:
            - {driver: nvidia, device_ids: ["1"], capabilities: [gpu]}

  worker-backend-1b:
    <<: *worker-backend-base
    container_name: minicpm-wb-1b
    volumes:
      - ${MODEL_HOST_PATH:?set MODEL_HOST_PATH to the host model dir}:/models/MiniCPM-o-4_5:ro
    deploy:
      resources:
        reservations:
          devices:
            - {driver: nvidia, device_ids: ["1"], capabilities: [gpu]}

  gateway:
    build:
      context: .
      dockerfile: docker/Dockerfile.gateway
    image: minicpm-gateway:dev
    container_name: minicpm-gateway
    command:
      - "--host"
      - "0.0.0.0"
      - "--port"
      - "8006"
      - "--https"
      - "--ssl-certfile"
      - "/app/certs/cert.pem"
      - "--ssl-keyfile"
      - "/app/certs/key.pem"
      - "--workers"
      - "worker-backend-0a:22400,worker-backend-0b:22400,worker-backend-1a:22400,worker-backend-1b:22400"
    ports:
      - "8006:8006"
    volumes:
      - ${DATA_HOST_PATH:-./data}:/app/data
      - ${CERTS_HOST_PATH:-./certs}:/app/certs:ro
    depends_on:
      worker-backend-0a: {condition: service_healthy}
      worker-backend-0b: {condition: service_healthy}
      worker-backend-1a: {condition: service_healthy}
      worker-backend-1b: {condition: service_healthy}
    restart: unless-stopped
    logging:
      driver: json-file
      options: {max-size: "50m", max-file: "3"}