user_turn_lora/docker-compose.yml at main · Prof-it/user_turn_lora · 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
version: '3.8'

services:
  userturn-lora:
    build: .
    image: userturn-lora:latest
    container_name: userturn-lora

    # GPU support
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

    # Mount volumes for data persistence
    volumes:
      - ./output:/app/output
      - ./data:/app/data
      - ~/.cache/huggingface:/root/.cache/huggingface

    # Environment variables
    environment:
      - HF_TOKEN=${HF_TOKEN}
      - WANDB_API_KEY=${WANDB_API_KEY}
      - HF_HUB_ENABLE_HF_TRANSFER=1
      - PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True,max_split_size_mb:64

    # Default command
    command: ["--model", "Qwen/Qwen2.5-3B-Instruct"]

  # Quick test run with smaller dataset using fastest model
  userturn-lora-test:
    build: .
    image: userturn-lora:latest
    container_name: userturn-lora-test

    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

    volumes:
      - ./output:/app/output
      - ~/.cache/huggingface:/root/.cache/huggingface

    environment:
      - HF_TOKEN=${HF_TOKEN}
      - WANDB_API_KEY=${WANDB_API_KEY}
      - HF_HUB_ENABLE_HF_TRANSFER=1

    command: [
      "--model", "LiquidAI/LFM2.5-1.2B-Instruct",
      "--train-samples", "100",
      "--eval-samples", "10",
      "--epochs", "1",
      "--no-wandb"
    ]