LLMDataForge/docker-compose.yml at master · sVarlog/LLMDataForge · 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
version: "3.9"

services:
    train:
        environment:
            - DISABLE_TRANSFORMERS_SDPA=1
            - PYTHONPATH=/workspace
            - RUN_MODE=train
        build:
            context: .
            dockerfile: Dockerfile
        image: llm-trainer
        container_name: llm-training
        volumes:
            - C:/Users/pc/.cache/huggingface:/root/.cache/huggingface
            - ./:/workspace
            - ./config:/workspace/config
        working_dir: /workspace
        command: python src/main.py
        deploy:
            resources:
                reservations:
                    devices:
                        - driver: nvidia
                          count: all
                          capabilities: [gpu]
        tty: true
        shm_size: "2gb"
        ulimits:
            memlock:
                soft: -1
                hard: -1
            stack: 67108864

    # test base + latest adapaters from checkpoint
    test-training:
        image: llm-trainer
        working_dir: /workspace
        command: python -u src/main.py
        container_name: llm-testing-trained
        environment:
            - PYTHONPATH=/workspace
            - RUN_MODE=test-training
            - TEST_MODE=force_think
        volumes:
            - C:/Users/pc/.cache/huggingface:/root/.cache/huggingface
            - ./:/workspace

    # test latest merged HF
    test-merging:
        image: llm-trainer
        working_dir: /workspace
        command: python -u src/main.py
        container_name: llm-testing-merged
        environment:
            - PYTHONPATH=/workspace
            - RUN_MODE=test-merging
            - TEST_MODE=force_think
        volumes:
            - ./:/workspace

    # test latest gguf under the latest merging-N
    test-gguf:
        image: llm-trainer
        working_dir: /workspace
        command: >
            bash -lc "pip install -q --no-cache-dir llama-cpp-python==0.3.5 && N_GPU_LAYERS=0 python -u src/main.py"
        container_name: llm-testing-gguf
        environment:
            - PYTHONPATH=/workspace
            - RUN_MODE=test-gguf
            - TEST_MODE=force_think
            - GGUF_CHAT_FORMAT=qwen
        volumes:
            - ./:/workspace
        deploy:
            resources:
                reservations:
                    devices:
                        - driver: nvidia
                          count: all
                          capabilities: [gpu]