Update Torch to ONNX export in conformance

[kshpv]

Changes

Do constant folding while exporting to ONNX from Torch

Reason for changes

Conformance test regressgion of ONNX after updating torch to 2.1
Model graphs are updated and contain BatchNorm. Therefore bias locates no more as Conv attribute but in BatchNorm layer. It leads to not applying FBC and BC algorithms to these biases.

Related tickets

125203

Tests

N/A

[codecov]

Codecov Report

Merging #2269 (1f7fef7) into develop (8efd04a) will increase coverage by 0.03%.
Report is 3 commits behind head on develop.
The diff coverage is n/a.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #2269      +/-   ##
===========================================
+ Coverage    90.72%   90.75%   +0.03%     
===========================================
  Files          485      485              
  Lines        43722    43715       -7     
===========================================
+ Hits         39665    39673       +8     
+ Misses        4057     4042      -15     

Flag	Coverage Δ
COMMON	15.82% <ø> (+<0.01%)	⬆️
ONNX	33.83% <ø> (+<0.01%)	⬆️
OPENVINO	38.62% <ø> (+0.04%)	⬆️
TENSORFLOW	30.04% <ø> (+<0.01%)	⬆️
TORCH	62.72% <ø> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

see 2 files with indirect coverage changes

[vshampor]

Since both representations are possible, shouldn't we instead modify the BC/FBC algos, or NNCFGraph creation from ONNX models to work correctly in both cases?

[kshpv]

Since both representations are possible, shouldn't we instead modify the BC/FBC algos, or NNCFGraph creation from ONNX models to work correctly in both cases?

Sure

[kshpv]

Conformance job 214. According to results ONNX problems were fixed. Also, some problems on develop was faced for Torch
@AlexanderDokuchaev please confirm

[AlexanderDokuchaev]

Conformance job 214. According to results ONNX problems were fixed. Also, some problems on develop was faced for Torch @AlexanderDokuchaev please confirm

Yes, ONNX works, but dont forget to update reference metrics. TORCH backend falls by know issue (ticker 125357)

[kshpv]

Conformance job 214. According to results ONNX problems were fixed. Also, some problems on develop was faced for Torch @AlexanderDokuchaev please confirm

Yes, ONNX works, but dont forget to update reference metrics. TORCH backend falls by know issue (ticker 125357)

Done

[alexsu52]

LGTM