* Introduce ONNX GraphConverter to get NNCFGraph from ONNX

* Place tests from experimental/onnx to onnx; Introduce ONNXGraph class instead ONNXGraphHelper; Some minor code changes;

* Fix pylint; Add requirements in tests/onnx

* Fix typo

* Add license, fix some comments

* minor

* Add docstring

* Update typehints

* Fix getting shapes of inputs, outputs; Add pylint exceptions

* Add requirements to ONNX

* Apply comments

* Add dtype attribute mapping from ONNX to NNCFGraph

* Align input, output nodes name with common NNCF; Extend onnx types to NNCF types mapping; Add test model with int edge

* Fix Vasiliy comment

* Add dot file

* Update requirements.tx

* Add hardware patterns; Add harware config support

* Create an early draft on PTQ API proposal

* Update API

* Place initialization flow to Initializer

* Add post-training quantization config

* Fix typos

* Add logic of framework agnostic building CompressedModel inside CompressionBuilder; Add serialization step;

* Implement the api ideas

* draft

* Implement algo with minimum inference of model

* draft x3

* draft x4

* Update sample

* Apply comments

* new draft

* Remove bn_adaptation files; Improve code style

* Fix some todos

* add draft biascorrection

* change directories

* update algorithms

* Improve code style

* Fix bug with adding q/dq to the model;
Now all models IR version are changed to version 7;
Add functional test on quantization

* Remove all extra code;

* Fix typo; remove queue;

* Remove ONNXUpdateBias and bias transformations; Delete comments + debug code;

* Add test on quantized models graphs;
Make input argument to ONNXEngine strictly np.ndarray;
Add statistics collection to apply method of ONNXQuantizerRangeFinderAlgorithm;
Make sampler working with torch and numpy;
Remove test of ptq sample;

* Change name QuantizerRangeFinderAlgorithm to MinMaxQuantization;
Some minor changes

* Hide torchvision imports in helper

* Fix codestyle

* Remove CompressedModel

* Add typehints;
Add small class descriptions

* Add test of graph after PostTrainingQuantization

* Translate string parameters to ENUM;
Minor changes

* Add ONNXMeanMinMaxStatisticCollector;
Fix bugs;
Set default range_type to MEAN_MINMAX;
Update requirements;

* Small improvements in Engine;

* Add test on ONNXModelTranformer;
Small improvements;

* Add many quantizers in one transformation layout for test_model_transformer;

* Improve codestyle

* Add test on parameters of inserted quantizers by ONNXModelTransformer

* Now algorithms don't collect statistics inside. They get statistics as an argument to the function apply(). So statistics always should run before the algorothms; Now CompressionBuilder is essential;

* Make batch_size=1;
The support should be done in the following PRs;
Minor changes

* Rename statistics_collector to statistics_aggregator

* Fix test

* Fix test x2;
Replace PTQ algo to algorithm.py

* Add license;
Create algorithm.py

* Fix pylint;
Add torch verion in requirements

* Add test_sampler;
Fix bug in BatchSampler and RandomBatchSampler;

* Add torchvision in requirements.txt;
Move docstring

* Improve codestyle

* Determine backend once;
Add create_subalgorithms() to Algorithm;
Rename test

* Rename files and functions;

* Make names Contants in ModelTransformer;
Add more comments;
Improve Codestyle

* Typo fix

Co-authored-by: Lyalyushkin Nikolay <[email protected]>

* Typo fix

Co-authored-by: Lyalyushkin Nikolay <[email protected]>

* Add test on StatisticsAggregator;
Fix bug in test_samplers.py

* Improve code style

* Change deque to List

* Add nncf_logger

* Move Constants inside ONNXModelTransformer

* Place min_max_quantization.py to quantization folder;
Fix tests

* Fix pylint

* Fix pylint;
Rename utils.py to model_normalizer.py

* Fix pylint

* Add ONNX ptq sample

* Add licences;
Add mock dataset to test_sanity_sample.py

* Add Readme;
Change dir;
Add AC configs

* Add results

* Update results table

* Rename and remove dataset path from AC configs;
Add sample requirements;
Improve readme

* Add mean, std, crop_ration into args of create_dataloader_from_imagenet_torch_dataset

* Fix readme

* Add description;
Remove 'train' postfix in creating dataset path

* Make default init_samples to 300;
Update metrics calibrated on val part of ImageNet

Co-authored-by: Lyalyushkin Nikolay <[email protected]>

Version update

modify networkx for CI testing