C++ コマンドラインインターフェイス

Nnabla は C++ バージョンのコマンドラインインターフェイスを持っており、学習や推論 ( フォワード ) を実行できます。このコマンドラインインターフェイスを使うことで、開発者は python 環境がなくても学習や推論を実行できます。

usage: nbla (infer|dump|train)

基本機能

Forward

usage: nbla infer -e EXECUTOR [-b BATCHSIZE] [-o OUTPUT] input_files ...

arguments:
   -e EXECUTOR         EXECUTOR is the name of executor network.
   input_files         input_file must be one of followings.
                           *.nnp      : Network structure and parameter.
                           *.nntxt    : Network structure in prototxt format.
                           *.prototxt : Same as nntxt.
                           *.h5       : Parameters in h5 format.
                           *.protobuf : Network structure and parameters in binary.
                           *.bin      : Input data.

optional arguments:
   -b BATCHSIZE        batch size for the input data.
   -o OUTPUT           the filename pattern of output file, default output to stdout.

example:
    Infer using LeNet_input.bin as input, LeNet_output_0.bin as output:
       nbla infer -e Executor -b 1 LeNet.nnp LeNet_input.bin -o LeNet_output

    Infer and output the result to console:
       nbla infer -e Executor -b 1 LeNet.nnp LeNet_input.bin

Dump

usage: nbla dump input_files ...

arguments:
   input_files         input_files must be one of *.nnp, *.nntxt, prototxt, h5, protobuf

example:
    Show network information by dump command:
      nbla dump LeNet.nnp

出力例 :

This configuration has 1 executors.

  Executor No.0 Name [Executor]
    Using default batch size 64 .
     Inputs
      Input No.0 Name [x] Shape ( 64 1 28 28 )
     Outputs
      Output No.0 Name [y'] Shape ( 64 10 )
Finished

学習

usage: nbla train input_file

arguments:
   input_file          input_file must be *.nnp