test-bench/docker-compose.yml at main · LLM-Dev-Ops/test-bench · 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
version: '3.8'

services:
  llm-test-bench:
    build:
      context: .
      dockerfile: Dockerfile
    image: llm-test-bench:latest
    container_name: llm-test-bench
    environment:
      # API Keys (set in .env file)
      - OPENAI_API_KEY=${OPENAI_API_KEY}
      - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}

      # Configuration
      - RUST_LOG=${RUST_LOG:-info}
      - LLM_TEST_BENCH_CONFIG=/data/config.toml

      # Evaluation settings
      - LLM_TEST_BENCH_EVALUATION__JUDGE_PROVIDER=${JUDGE_PROVIDER:-openai}
      - LLM_TEST_BENCH_EVALUATION__JUDGE_MODEL=${JUDGE_MODEL:-gpt-4}
      - LLM_TEST_BENCH_EVALUATION__CACHE_ENABLED=${CACHE_ENABLED:-true}

    volumes:
      # Mount datasets
      - ./datasets:/data/datasets:ro

      # Mount results directory
      - ./results:/data/results:rw

      # Mount cache directory
      - llm-cache:/data/cache:rw

      # Mount configuration (optional)
      - ./config.toml:/data/config.toml:ro

    # Override command for specific operations
    # Uncomment and modify as needed:

    # Run benchmark
    # command: >
    #   bench
    #   --dataset /data/datasets/coding-tasks.json
    #   --providers openai,anthropic
    #   --metrics faithfulness,relevance
    #   --output /data/results
    #   --dashboard

    # Compare models
    # command: >
    #   compare
    #   --prompt "Explain quantum computing"
    #   --models openai:gpt-4,anthropic:claude-3-opus
    #   --statistical-tests
    #   --output /data/results/comparison.html

    # Generate dashboard
    # command: >
    #   dashboard
    #   --results /data/results/*.json
    #   --theme dark
    #   --output /data/results/dashboard.html

    # Analyze performance
    # command: >
    #   analyze
    #   --baseline /data/results/baseline.json
    #   --comparison /data/results/latest.json
    #   --fail-on-regression

    # Optimize costs
    # command: >
    #   optimize
    #   --current-model gpt-4
    #   --monthly-requests 100000
    #   --quality-threshold 0.80

    # Resource limits
    deploy:
      resources:
        limits:
          cpus: '2.0'
          memory: 2G
        reservations:
          cpus: '0.5'
          memory: 512M

    # Restart policy
    restart: unless-stopped

    # Networking
    networks:
      - llm-network

volumes:
  llm-cache:
    driver: local

networks:
  llm-network:
    driver: bridge