######################
Automatic Quantization
######################

``enot.quantization`` package contains functional for automatic quantization
of user models. Best suitable for preparing user models for `enot-lite`_ int8
engines.

With ``enot.quantization`` package, you can automatically convert your
``PyTorch`` model to our intermediate representation which allows you to
perform multiple kinds of quantization including vector quantization for
`TensorRT`_ and `OpenVINO`_.

This package features automatic distillation for weight fine-tuning, automatic
quantization threshold search described in `Fast Adjustable Threshold`_ paper,
different methods for layer selection for distillation and a number of
fake-quantization algorithms.

.. _enot-lite: https://enot-lite.rtd.enot.ai/en/stable/
.. _OpenVINO: https://docs.openvino.ai/latest/index.html
.. _TensorRT: https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html
.. _Fast Adjustable Threshold: https://arxiv.org/abs/1812.07872


**********************
enot-lite quantization
**********************

.. autoclass:: enot.quantization.TrtFakeQuantizedModel
   :members: __init__, quantization_parameters, regular_parameters, enable_calibration_mode, enable_quantization_mode
   :show-inheritance:

.. autoclass:: enot.quantization.OpenvinoFakeQuantizedModel
   :members: __init__, quantization_parameters, regular_parameters, enable_calibration_mode, enable_quantization_mode
   :show-inheritance:

.. autoclass:: enot.quantization.FakeQuantizedModel
   :members: __init__, quantization_parameters, regular_parameters, enable_calibration_mode, enable_quantization_mode
   :show-inheritance:

.. autoclass:: enot.quantization.utils.QuantizationGranularity
   :members:
   :undoc-members:

.. autoclass:: enot.quantization.utils.QuantizationStrategy
   :members:
   :undoc-members:

.. autoclass:: enot.quantization.utils.CalibrationMethod
   :members:
   :undoc-members:

.. autoclass:: enot.quantization.utils.RoundingFunction
   :members:
   :undoc-members:

.. autoclass:: enot.quantization.utils.QuantizationType
   :members:

.. autofunction:: enot.quantization.float_model_from_quantized_model

.. autofunction:: enot.quantization.utils.optimal_quantization_scheme


***********
calibration
***********

Functions and classes from ``calibration`` module provides easy functional to
calibrate quantization thresholds in fake-quantized model.

.. autofunction:: enot.quantization.calibrate_quantized_model

.. autoclass:: enot.quantization.calibration_context
    :members: __init__
    :show-inheritance:


************
distillation
************

Functions and classes from ``distillation`` module provides utilities and
procedures for quantized model fine-tuning using `knowledge distillation`_.

.. _knowledge distillation: https://arxiv.org/abs/1503.02531

Helper functions for distillation:

.. autoclass:: enot.distillation.DistillationLayer
    :show-inheritance:

.. autoclass:: enot.quantization.QuantDistillationModule
    :show-inheritance:

.. autoclass:: enot.quantization.DistillationLayerSelectionStrategy
    :show-inheritance:

.. autofunction:: enot.quantization.add_distillation_nodes_to_onnx_converted_model

.. autoclass:: enot.quantization.distillation_context
    :show-inheritance:

We are also open-sourcing our distillation procedures to allow user to
customize distillation procedure. All classes below are public, and you can
view their source code by clicking on ``[source]`` links.

.. autoclass:: enot.quantization.RMSELoss
    :show-inheritance:

.. autoclass:: enot.quantization.DistillerInterface

.. autoclass:: enot.quantization.QuantizationDistiller
    :show-inheritance:
    :members:

.. autoclass:: enot.quantization.SequentialDistiller
    :show-inheritance:
    :members:

.. autoclass:: enot.quantization.DefaultQuantizationDistiller
    :show-inheritance:

.. autoclass:: enot.quantization.ThresholdsAndScaleFactorsQuantizationDistiller
    :show-inheritance:
    :members:

.. autoclass:: enot.quantization.ThresholdsQuantizationDistiller
    :show-inheritance:
    :members: