Issues when importing to C++

[alkaf-ahmed]

I'm having trouble when importing a simple DNN for MNIST to C++

Here's the content of the .ini file

DefaultModel=Frame_CUDA

; Database
[database]
Type=MNIST_IDX_Database
Validation=0.2 ; Use 20% of the dataset for validation

; Environment
[sp]
SizeX=32
SizeY=32
BatchSize=128

[sp.Transformation_1]
Type=RescaleTransformation
Width=[sp]SizeX
Height=[sp]SizeY

[fc1]
Input=sp
Type=Fc
NbOutputs=84
ActivationFunction=Rectifier
WeightsFiller=HeFiller
ConfigSection=common.config

[fc2]
Input=fc1
Type=Fc
NbOutputs=84
ActivationFunction=Rectifier
WeightsFiller=HeFiller
ConfigSection=common.config

[fc3]
Input=fc2
Type=Fc
NbOutputs=84
ActivationFunction=Rectifier
WeightsFiller=HeFiller
ConfigSection=common.config

[fc4]
Input=fc3
Type=Fc
NbOutputs=10
ActivationFunction=Linear
ConfigSection=common.config

[softmax]
Input=fc4
Type=Softmax
NbOutputs=[fc4]NbOutputs
WithLoss=1

[softmax.Target]

[common.config]
NoBias=1
WeightsSolver.LearningRate=0.05
WeightsSolver.Momentum=0.9
WeightsSolver.Decay=0.0005
Solvers.LearningRatePolicy=StepDecay
Solvers.LearningRateStepSize=[sp]_EpochSize
Solvers.LearningRateDecay=0.993
Solvers.Clamping=-1.0:1.0

Here's the output after quantizing and calibration when executing the following command : n2d2 conf.ini -seed 1 -w /dev/null -export CPP -calib -1 -nbbits 8

Output :

Quantizing free parameters:
  - fc1: 0.324863
  - fc2: 0.138924
  - fc3: 0.0325353
  - fc4: 0.00761959
  - softmax: 0.00761959
  Fuse scaling cells:
  Quantizing activations:
  - fc1: prev=1, act=3.67446, bias=0.324863
      quant=127, global scaling=11.3108 -> cell scaling=0.000696151
  - fc2: prev=11.3108, act=2.49301, bias=0.138924
      quant=127, global scaling=17.9452 -> cell scaling=0.00496297
  - fc3: prev=17.9452, act=0.993608, bias=0.0325353
      quant=127, global scaling=30.5394 -> cell scaling=0.00462683
  - fc4: prev=30.5394, act=0.546552, bias=0.00761959
      quant=255, global scaling=71.7299 -> cell scaling=0.00166963
  Fuse scaling cells:
  - fuse: fc1_rescale_act
  - fuse: fc2_rescale_act
  - fuse: fc3_rescale_act
  - fuse: fc4_rescale_act
  Scaling approximation [1]:
  - fc1: 0.000696151
  - fc2: 0.00496297
  - fc3: 0.00462683
  - fc4: 0.00166963
  Inputs quantization
  Done!
sh: 1: dot: not found
Exporting Test dataset to "export_CPP_int8/stimuli"....................
Generating CPP export to "export_CPP_int8":
-> Generating network
Fuse Padding...
sh: 1: dot: not found
-> Generating cell fc1
-> Generating cell fc2
-> Generating cell fc3
-> Generating cell fc4
-> Generating cell softmax
Invalid registrar key "Softmax"
Notice: Unused section  in INI file
Time elapsed: 6.2989 s
Error: bad_function_call

Any ideas ?

Thanks in advance