Fix the logic how number of calibration batches are computed in PTQ t…

…o match the documentation (#1918)

* Removing pytorch_quantization, replacing with lazy install

* Removing pytorch_quantization, replacing with lazy install

* Reorganize imports

* Reorganize imports

* Fix imports

* Fix imports

* Fix imports

* import_onnx_graphsurgeon_or_install

* import_onnx_graphsurgeon_or_install

* Remove TODOs

* Remove TODOs

* Fix imports

* Fix install command

* Merge master

* Add install

* update notebbok

* Update test

* use_fb_fake_quant

* Simplify export methods

* Simplify export methods

* Simplify export methods

* Fix use_fb_fake_quant

* Fix typo

* Make sure that we follow documentation on selecting number of calibration batches

* Remove extra call to quantize_module

* Set model to eval before PTQ

* Print quantization results to console

* Added tqdm descirption

* Set default calibration batches to 16

* Re-order arguments

* Re-run notebook

---------

Co-authored-by: Shay Aharon <[email protected]>