Frameworks supported by TinyMind are:

  • TensorFlow (1.2, 1.3, 1.4)
  • Keras (2.0, with TensorFlow or Theano backend)
  • Caffe2 (0.8)
  • Theano (0.9)
  • MXNet (0.11)
  • PyTorch (0.1, 0.2, 0.3)

All frameworks are available in both CPU and GPU versions. All frameworks other than Caffe2 support Python 2.7, 3.5 and 3.6. Caffe2 currently only supports Python 2.7.

TinyMind maintains performance-optimized builds of the frameworks. Patch versions are automatically made available as they are released. You will need to manually upgrade your models to new major or minor versions as they become available.


Latest versions of the following dependencies are by default available to your model:

  • h5py
  • matplotlib
  • mkl
  • numpy
  • pandas
  • pillow
  • pygpu (for GPU environments)
  • pyyaml
  • scikit-learn
  • scipy
  • six
  • tinyenv

If your code archive or GitHub repo has a requirements.txt file at its root level, dependencies in it will be automatically installed. You can also input custom dependencies into the Dependencies text field, like such:

Custom Dependencies