Customizing Environments

Environments

Spell provides a number of environments for you to run your code in. All environments have the Ubuntu 18.04 operating system, Python versions 2.7 and 3.6, xgboost, and scikit-learn installed. If a machine type is specified that has a GPU, then CUDA and cuDNN are also included.

There are a number of open-source frameworks that are used for deep learning. Spell supports most of them.

Note

If you don't see your preferred framework, please request it by emailing us at support@spell.run or leaving us feedback using the spell feedback command in our CLI.

Supported Environments

Name CLI arguments Notes
Default (none) tensorflow==1.14.0
torch==1.2.0
torchvision==0.4.0
Pillow==6.1.0
Keras==2.2.4
MXNet --framework mxnet mxnet==1.5.0
fast.ai --framework fastai torch==1.2.0
Caffe --framework caffe caffe==1.0.0
DyNet --framework dynet dyNET==2.1
Torch --framework torch Torch7
Conda --conda-file [path-to-conda-file] Using a conda environment file or spec file.

Managing Dependencies

Learn about how to add pip, apt, conda and other dependencies here.

Pip Packages Installed by Default

Every framework comes with the following default pip packages already installed:

six==1.12.0
scipy==1.2.1
requests==2.21.0
protobuf==3.6.1
pickleshare==0.7.5
numpy==1.16.2
h5py==2.9.0