localInstallGuide/docker-compose.yml at main · erenakyildiz-code/localInstallGuide · 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
# Linux / Windows (WSL2) with an NVIDIA GPU.
# Usage:  cp .env.example .env   (edit it)   →   docker compose up -d
name: companion

services:
  # ── BRAIN: llama.cpp server, OpenAI-ish API on :8080 ──────────────────────
  brain:
    build:
      context: .
      dockerfile: Dockerfile.brain
    container_name: companion-brain
    restart: unless-stopped
    ports:
      - "127.0.0.1:8080:8080"
    volumes:
      - ./GGUF:/models          # put your .gguf files in ./GGUF
    ulimits:
      memlock:
        soft: -1
        hard: -1
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all        # or device_ids: ['0'] to pin one GPU
              capabilities: [gpu, compute, utility]
    command: [
      "-m", "/models/${BRAIN_MODEL}",
      "--host", "0.0.0.0",
      "--port", "8080",
      "-c", "${BRAIN_CTX}",          # context size — 26B Q6 at 171072 needs a 24GB+ card
      "-b", "8192",
      "-ngl", "99",                  # layers offloaded to GPU (99 = all)
      "--flash-attn", "on",
      "--cache-type-k", "q8_0",      # KV cache quant — big VRAM saver on long context
      "--cache-type-v", "q8_0",
      "--chat-template", "gemma"
    ]

  # ── EARS: faster-whisper ASR, OpenAI-compatible /v1/audio/transcriptions ──
  ears:
    image: fedirz/faster-whisper-server:latest-cuda
    container_name: companion-ears
    restart: unless-stopped
    ports:
      - "127.0.0.1:8000:8000"
    environment:
      - WHISPER__MODEL=${WHISPER_MODEL}
      - WHISPER__INFERENCE_DEVICE=cuda
    volumes:
      - ears-models:/root/.cache/huggingface   # model survives container rebuilds
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

  # ── MOUTH: Kokoro TTS, OpenAI-compatible /v1/audio/speech ─────────────────
  mouth:
    image: ghcr.io/remsky/kokoro-fastapi-gpu:latest
    container_name: companion-mouth
    restart: unless-stopped
    ports:
      - "127.0.0.1:8001:8880"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

volumes:
  ears-models: