longkeyy/llamafactory Docker 镜像 - 轩辕镜像 | Docker 镜像高效稳定拉取服务

LlamaFactory Docker镜像使用指南

longkeyy/llamafactory是一个多功能Docker镜像，支持多种硬件后端的模型训练与推理，包括NVIDIA CUDA、华为Ascend NPU和AMD ROCm。它提供Web UI和OpenAI兼容的API访问方式。

可用标签与环境

注意： latest标签始终指向最新的cuda镜像版本。

快速开始

CUDA版本（NVIDIA GPU）

bash
docker run -it --gpus all \
  --shm-size 16g \
  --ipc host \
  -p 7860:7860 -p 8000:8000 \
  -v $PWD/hf_cache:/root/.cache/huggingface \
  -v $PWD/ms_cache:/root/.cache/modelscope \
  -v $PWD/om_cache:/root/.cache/openmind \
  -v $PWD/data:/app/data \
  -v $PWD/output:/app/output \
  --restart unless-stopped \
  longkeyy/llamafactory:cuda-0.9.2

NPU版本（华为Ascend）

确保已挂载必要的Ascend驱动和设备。

bash
docker run -it \
  --shm-size 16g \
  --ipc host \
  --device /dev/davinci0 \
  --device /dev/davinci_manager \
  --device /dev/devmm_svm \
  --device /dev/hisi_hdc \
  -v /usr/local/dcmi:/usr/local/dcmi \
  -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
  -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
  -v /etc/ascend_install.info:/etc/ascend_install.info \
  -v $PWD/hf_cache:/root/.cache/huggingface \
  -v $PWD/ms_cache:/root/.cache/modelscope \
  -v $PWD/om_cache:/root/.cache/openmind \
  -v $PWD/data:/app/data \
  -v $PWD/output:/app/output \
  -p 7860:7860 -p 8000:8000 \
  longkeyy/llamafactory:npu-0.9.2

ROCm版本（AMD GPU）

bash
docker run -it \
  --shm-size 16g \
  --ipc host \
  --device /dev/kfd \
  --device /dev/dri \
  -v $PWD/hf_cache:/root/.cache/huggingface \
  -v $PWD/ms_cache:/root/.cache/modelscope \
  -v $PWD/om_cache:/root/.cache/openmind \
  -v $PWD/data:/app/data \
  -v $PWD/output:/app/output \
  -v $PWD/saves:/app/saves \
  -p 7860:7860 -p 8000:8000 \
  --restart unless-stopped \
  longkeyy/llamafactory:rocm-0.9.2

Docker Compose配置

docker-compose.yml（多后端支持）

yaml
version: '3.8'

services:
  llamafactory-cuda:
    image: longkeyy/llamafactory:cuda-0.9.2
    container_name: llamafactory-cuda
    ports:
      - "7860:7860"
      - "8000:8000"
    volumes:
      - ./hf_cache:/root/.cache/huggingface
      - ./ms_cache:/root/.cache/modelscope
      - ./om_cache:/root/.cache/openmind
      - ./data:/app/data
      - ./output:/app/output
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    shm_size: "16gb"
    ipc: host
    tty: true
    restart: unless-stopped
    profiles: ["cuda"]

  llamafactory-npu:
    image: longkeyy/llamafactory:npu-0.9.2
    container_name: llamafactory-npu
    ports:
      - "7860:7860"
      - "8000:8000"
    volumes:
      - ./hf_cache:/root/.cache/huggingface
      - ./ms_cache:/root/.cache/modelscope
      - ./om_cache:/root/.cache/openmind
      - ./data:/app/data
      - ./output:/app/output
      - /usr/local/dcmi:/usr/local/dcmi
      - /usr/local/bin/npu-smi:/usr/local/bin/npu-smi
      - /usr/local/Ascend/driver:/usr/local/Ascend/driver
      - /etc/ascend_install.info:/etc/ascend_install.info
    devices:
      - /dev/davinci0
      - /dev/davinci_manager
      - /dev/devmm_svm
      - /dev/hisi_hdc
    shm_size: "16gb"
    ipc: host
    tty: true
    stdin_open: true
    restart: unless-stopped
    profiles: ["npu"]

  llamafactory-rocm:
    image: longkeyy/llamafactory:rocm-0.9.2
    container_name: llamafactory-rocm
    ports:
      - "7860:7860"
      - "8000:8000"
    volumes:
      - ./hf_cache:/root/.cache/huggingface
      - ./ms_cache:/root/.cache/modelscope
      - ./om_cache:/root/.cache/openmind
      - ./data:/app/data
      - ./output:/app/output
      - ./saves:/app/saves
    devices:
      - /dev/kfd:/dev/kfd
      - /dev/dri:/dev/dri
    shm_size: "16gb"
    ipc: host
    tty: true
    stdin_open: true
    restart: unless-stopped
    profiles: ["rocm"]

指定后端运行

bash
# CUDA
docker compose --profile cuda up -d

# NPU
docker compose --profile npu up -d

# ROCm
docker compose --profile rocm up -d

挂载卷说明

访问地址

• Web UI: http://localhost:7860
• API端点: http://localhost:8000

最佳实践

• 使用--restart unless-stopped参数确保服务弹性。
• 镜像默认命令为llamafactory-cli webui。
• 如需扩展功能，可通过Dockerfile的ARG参数自定义构建。

镜像大小概览