[PT] Extractor for fused convolution

[AlexanderDokuchaev]

Changes

• Get layer attributes by args, kwargs for conv, batch_norm, group_norm and linear functions.
• Extract submodules for FBC for custom conv and batch norm modules.
• Add model_graph_manager module to works with models with wrapped with trace_parameters=True. (model_analizer.py will be deprecated after finished refactor PTQ to use graph with constant nodes)
• Moved nncf/quantization/algorithms/weight_compression/torch_backend.py to model_graph_manager

Related tickets

129581

Tests

tests/torch/test_model_graph_manager.py
tests/torch/test_extractor.py

[codecov]

Codecov Report

Attention: Patch coverage is 89.26380% with 35 lines in your changes are missing coverage. Please review.

Project coverage is 84.84%. Comparing base (c5b3df8) to head (906800f).
Report is 27 commits behind head on develop.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #2559      +/-   ##
===========================================
- Coverage    90.92%   84.84%   -6.08%     
===========================================
  Files          498      494       -4     
  Lines        45550    45332     -218     
===========================================
- Hits         41415    38464    -2951     
- Misses        4135     6868    +2733     

Files	Coverage Δ
nncf/common/graph/layer_attributes.py	96.29% <100.00%> (+0.03%)	⬆️
nncf/common/graph/operator_metatypes.py	93.97% <100.00%> (+0.14%)	⬆️
...ion/algorithms/weight_compression/torch_backend.py	84.00% <100.00%> (-1.09%)	⬇️
nncf/torch/dynamic_graph/wrappers.py	96.55% <100.00%> (+0.44%)	⬆️
nncf/torch/graph/operator_metatypes.py	99.10% <100.00%> (-0.13%)	⬇️
...f/torch/dynamic_graph/layer_attributes_handlers.py	99.36% <98.63%> (-0.64%)	⬇️
nncf/torch/extractor.py	83.95% <83.95%> (ø)
nncf/torch/model_graph_manager.py	84.21% <84.21%> (ø)

... and 83 files with indirect coverage changes

Flag	Coverage Δ
COMMON	44.03% <18.75%> (+1.30%)	⬆️
ONNX	34.54% <14.72%> (-0.23%)	⬇️
OPENVINO	∅ <ø> (∅)
TENSORFLOW	29.72% <0.92%> (+0.06%)	⬆️
TORCH	65.87% <89.26%> (-0.14%)	⬇️

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

Components	Coverage Δ
common	93.11% <100.00%> (-0.64%)	⬇️
torch	93.48% <89.13%> (+0.04%)	⬆️
tensorflow	93.74% <ø> (-0.22%)	⬇️
onnx	93.05% <ø> (ø)
openvino	25.58% <ø> (-68.49%)	⬇️
ptq	69.67% <100.00%> (-18.07%)	⬇️

[daniil-lyakhov]

Looks good overall, please address my comments

[alexsu52]

Please, provide e2e results for PyTorch backend.

[AlexanderDokuchaev]

@alexsu52 TriggerBetta/647

[alexsu52]

LGTM