Deep Learning Containers (DLC) allows you to use public images or custom images to run deep learning jobs.

public image

DLC provides a variety of public images that support different types of resources, Python versions, and deep learning frameworks (TensorFlow and PyTorch). The following table lists the supported public images.

Variable Description Valid value
region Endpoints of supported regions
  • cn-hangzhou: China (Hangzhou)
  • cn-shanghai: China (Shanghai)
  • cn-beijing: China (Beijing)
  • cn-shenzhen: China (Shenzhen)
type Supported resource types
  • cpu
  • gpu
  • mkl_cpu: Math Kernel Library (MKI-DNN)
python Supported Python versions
  • py2: Python 2
  • py3: Python 3
The following lists show the types of resources and Python versions supported by different deep learning frameworks:
Note When you use an image, you must replace ${Variable} with the corresponding value listed in the preceding table.
  • TensorFlow 1.12 (CUDA version: nvidia-cuda-10.0-cudnn7)
    • Supported regions: China (Hangzhou), China (Shanghai), China (Beijing), and China (Shenzhen)
    • Supported Python versions: Python 2 and Python 3
    • Supported resource types: CPU, GPU, and MKI_CPU
    • Image path: registry.${region}.aliyuncs.com/pai-dlc/pai-tensorflow-training:1.12-${type}-${python}
    • Example: registry.cn-beijing.aliyuncs.com/pai-dlc/pai-tensorflow-training:1.12-cpu-py2

      In this example, TensorFlow 1.12, China (Beijing), CPU, and Python 2 are specified.

  • TensorFlow 1.15 (CUDA version: nvidia-cuda-10.0-cudnn7)
    • Supported regions: China (Hangzhou), China (Shanghai), China (Beijing), and China (Shenzhen)
    • Supported Python versions: Python 2 and Python 3
    • Supported resource types: GPU
    • Image path: registry.${region}.aliyuncs.com/pai-dlc/pai-tensorflow-training:1.15-${type}-${python}
    • Example: registry.cn-beijing.aliyuncs.com/pai-dlc/pai-tensorflow-training:1.15-gpu-py2

      In this example, TensorFlow 1.15, China (Beijing), GPU, and Python 2 are specified.

  • TensorFlow 2.2 (CUDA version: nvidia-cuda-10.1-cudnn7)
    • Supported regions: China (Hangzhou), China (Shanghai), China (Beijing), and China (Shenzhen)
    • Supported Python versions: Python 3
    • Supported resource types: CPU and GPU
    • Image path: registry.${region}.aliyuncs.com/pai-dlc/pai-tensorflow-training:2.2-${type}-${python}
    • Example: registry.cn-beijing.aliyuncs.com/pai-dlc/pai-tensorflow-training:2.2-cpu-py3

      In this example, TensorFlow 2.2, China (Beijing), CPU, and Python 3 are specified.

  • PyTorch 1.3 (CUDA version: nvidia-cuda-10.0-cudnn)
    • Supported regions: China (Hangzhou), China (Shanghai), China (Beijing), and China (Shenzhen)
    • Supported Python versions: Python 2 and Python 3
    • Supported resource types: CPU, GPU, and MKI_CPU
    • Image path: registry.${region}.aliyuncs.com/pai-dlc/pai-pytorch-training:1.3-${type}-${python}
    • Example: registry.cn-beijing.aliyuncs.com/pai-dlc/pai-pytorch-training:1.3-gpu-py3

      In this example, PyTorch 1.3, China (Beijing), GPU, and Python 3 are specified.

  • PyTorch 1.5 (CUDA version: nvidia-cuda-10.1-cudnn7)
    • Supported regions: China (Hangzhou), China (Shanghai), China (Beijing), and China (Shenzhen)
    • Supported Python versions: Python 2 and Python 3
    • Supported resource types: CPU, GPU, and MKI_CPU
    • Image path: registry.${region}.aliyuncs.com/pai-dlc/pai-pytorch-training:1.5-${type}-${python}
    • Example: registry.cn-beijing.aliyuncs.com/pai-dlc/pai-pytorch-training:1.5-gpu-py3

      In this example, PyTorch 1.5, China (Beijing), GPU, and Python 3 are specified.

Custom images

DLC allows you to use custom images. For more information about how to pull a custom image, see Basic operations on Docker images. The basic Docker image provided by DLC is: reg.docker.alibaba-inc.com/dlc/mirror:nvidia-cuda-10.0-cudnn7-devel-ubuntu16.04.