Current Support

MLModelScope currently

  • supports Caffe, Caffe2, CNTK, MXNet, PyTorch, TensorFlow and TensorRT
  • runs on ARM, PowerPC, and X86 with CPU, GPU, and FPGA
  • contains common vision models and datasets
  • has built-in framework, library and system profilers.

MLModelScope allows users to easily extend it by adding models, datasets, frameworks, library or system profilers, and systems. Refer to the Extending section for details.

Frameworks and Models

LModelScope has the following frameworks built in:

MLModelScope specifies models using “manifest”, and has the following models built into each framework predictor:


MLModelScope has been tested on the following hardware:

  • X86
  • PowerPC
  • ARM
  • GPU
  • FPGA


MLModelScope has the following datasets built in:

  • ilsvrc2012
  • cifar10, cifar100
  • caltech256
  • mnist

The datasets are stored in an efficient format that allows fast sequential reads from disk. During evaluation, this fast format is used to read the data.

Refer to dldataset for details.


MLModelScope has been tested on the following operating systems:

  • Linux
  • MacOS