[Runtime] INT8 inference documentation update

docs/IE_DG/Int8Inference.md

@@ -1,6 +1,6 @@
 # Low-Precision 8-bit Integer Inference {#openvino_docs_IE_DG_Int8Inference}
 
-## Disclaimer
+## Supported devices
 
 Low-precision 8-bit inference is optimized for:
 - Intel® architecture processors with the following instruction set architecture extensions:  
@@ -12,15 +12,22 @@ Low-precision 8-bit inference is optimized for:
   - Intel® Iris® Xe Graphics
   - Intel® Iris® Xe MAX Graphics
 - A model must be quantized. You can use a quantized model from [OpenVINO™ Toolkit Intel's Pre-Trained Models](@ref omz_models_group_intel) or quantize a model yourself. For quantization, you can use the:
-  - [Post-Training Optimization Tool](@ref pot_README) delivered with the Intel® Distribution of OpenVINO™ toolkit release package.
+  - [Post-Training Optimization Tool](@ref pot_docs_LowPrecisionOptimizationGuide) delivered with the Intel® Distribution of OpenVINO™ toolkit release package.
   - [Neural Network Compression Framework](https://www.intel.com/content/www/us/en/artificial-intelligence/posts/openvino-nncf.html) available on GitHub: https://github.com/openvinotoolkit/nncf
 
-## Introduction
+## Low-Precision 8-bit Integer Inference Workflow
+
+8-bit computation (referred to as `int8`) offers better performance compared to the results of inference in higher precision (for example, `fp32`), because it allows loading more data into a single processor instruction. Usually the cost for significant boost is reduced accuracy. However, it has been proved that the drop in accuracy can be negligible and depends on task requirements, so that the application engineer configures the maximum accuracy drop that is acceptable.
+
+For 8-bit integer computations, a model must be quantized. Quantized models can be downloaded from [Overview of OpenVINO™ Toolkit Intel's Pre-Trained Models](@ref omz_models_group_intel). If the model is not quantized, you can use the [Post-Training Optimization Tool](@ref pot_README) to quantize the model. The quantization process adds [FakeQuantize](../ops/quantization/FakeQuantize_1.md) layers on activations and weights for most layers. Read more about mathematical computations in the [Uniform Quantization with Fine-Tuning](https://github.com/openvinotoolkit/nncf/blob/develop/docs/compression_algorithms/Quantization.md).
 
-A lot of investigation was made in the field of deep learning with the idea of using low-precision computation during inference in order to boost deep learning pipelines and achieve higher performance. For example, one of the popular approaches is to shrink the precision of activations and weights values from `fp32` precision to smaller ones, for example, to `fp11` or `int8`. For more information about this approach, refer to the
-**Brief History of Lower Precision in Deep Learning** section in [this whitepaper](https://software.intel.com/en-us/articles/lower-numerical-precision-deep-learning-inference-and-training).
+When you pass the quantized IR to the OpenVINO™ plugin, the plugin automatically recognizes it as a quantized model and performs 8-bit inference. Note, if you pass a quantized model to another plugin that does not support 8-bit inference but supports all operations from the model, the model is inferred in precision that this plugin supports.
 
-8-bit computation (referred to as `int8`) offers better performance compared to the results of inference in higher precision (for example, `fp32`), because they allow loading more data into a single processor instruction. Usually the cost for significant boost is reduced accuracy. However, it has been proven that the drop in accuracy can be negligible and depends on task requirements, so that an application engineer configure the maximum accuracy drop that is acceptable.
+In *Runtime stage*, the quantized model is loaded to the plugin. The plugin uses the `Low Precision Transformation` component to update the model to infer it in low precision:
+   - Update `FakeQuantize` layers to have quantized output tensors in a low precision range and add dequantization layers to compensate the update. Dequantization layers are pushed through as many layers as possible to have more layers in low precision. After that, most layers quantized input tensors in the low precision range and can be inferred in low precision. Ideally, dequantization layers should be fused in the next `FakeQuantize` layer.
+   - Quantize weights and store them in `Constant` layers. 
+
+## Prerequisites
 
 Let's explore the quantized [TensorFlow* implementation of ResNet-50](https://github.com/openvinotoolkit/open_model_zoo/tree/master/models/public/resnet-50-tf) model. Use the [Model Downloader](@ref omz_tools_downloader) tool to download the `fp16` model from [OpenVINO™ Toolkit - Open Model Zoo repository](https://github.com/openvinotoolkit/open_model_zoo):
 ```sh
@@ -31,28 +38,16 @@ After that, you should quantize the model by the [Model Quantizer](@ref omz_tool
 ```sh
 ./quantizer.py --model_dir --name public/resnet-50-tf --dataset_dir <DATASET_DIR> --precisions=FP16-INT8
 ```
+
+## Inference
+
 The simplest way to infer the model and collect performance counters is the [C++ Benchmark Application](../../inference-engine/samples/benchmark_app/README.md). 
 ```sh
 ./benchmark_app -m resnet-50-tf.xml -d CPU -niter 1 -api sync -report_type average_counters  -report_folder pc_report_dir
 ```
 If you infer the model with the OpenVINO™ CPU plugin and collect performance counters, all operations (except the last non-quantized SoftMax) are executed in INT8 precision.  
 
-## Low-Precision 8-bit Integer Inference Workflow
-
-For 8-bit integer computations, a model must be quantized. Quantized models can be downloaded from [Overview of OpenVINO™ Toolkit Intel's Pre-Trained Models](@ref omz_models_group_intel). If the model is not quantized, you can use the [Post-Training Optimization Tool](@ref pot_README) to quantize the model. The quantization process adds [FakeQuantize](../ops/quantization/FakeQuantize_1.md) layers on activations and weights for most layers. Read more about mathematical computations in the [Uniform Quantization with Fine-Tuning](https://github.com/openvinotoolkit/nncf/blob/develop/docs/compression_algorithms/Quantization.md).
-
-8-bit inference pipeline includes two stages (also refer to the figure below):
-1. *Offline stage*, or *model quantization*. During this stage, [FakeQuantize](../ops/quantization/FakeQuantize_1.md) layers are added before most layers to have quantized tensors before layers in a way that low-precision accuracy drop for 8-bit integer inference satisfies the specified threshold. The output of this stage is a quantized model. Quantized model precision is not changed, quantized tensors are in original precision range (`fp32`). `FakeQuantize` layer has `levels` attribute which defines quants count. Quants count defines precision which is used during inference. For `int8` range `levels` attribute value has to be 255 or 256. To quantize the model, you can use the [Post-Training Optimization Tool](@ref pot_README) delivered with the Intel® Distribution of OpenVINO™ toolkit release package.
-
-   When you pass the quantized IR to the OpenVINO™ plugin, the plugin automatically recognizes it as a quantized model and performs 8-bit inference. Note, if you pass a quantized model to another plugin that does not support 8-bit inference but supports all operations from the model, the model is inferred in precision that this plugin supports.
-
-2. *Runtime stage*. This stage is an internal procedure of the OpenVINO™ plugin. During this stage, the quantized model is loaded to the plugin. The plugin uses `Low Precision Transformation` component to update the model to infer it in low precision:
-   - Update `FakeQuantize` layers to have quantized output tensors in low precision range and add dequantization layers to compensate the update. Dequantization layers are pushed through as many layers as possible to have more layers in low precision. After that, most layers have quantized input tensors in low precision range and can be inferred in low precision. Ideally, dequantization layers should be fused in the next `FakeQuantize` layer.
-   - Weights are quantized and stored in `Constant` layers. 
-
-![int8_flow]
-
-## Performance Counters
+## Results analysis
 
 Information about layer precision is stored in the performance counters that are
 available from the Inference Engine API. For example, the part of performance counters table for quantized [TensorFlow* implementation of ResNet-50](https://github.com/openvinotoolkit/open_model_zoo/tree/master/models/public/resnet-50-tf) model inference on [CPU Plugin](supported_plugins/CPU.md) looks as follows:
@@ -79,5 +74,3 @@ available from the Inference Engine API. For example, the part of performance co
 > * Suffix `FP32` for layers computed in 32-bit precision 
 
 All `Convolution` layers are executed in int8 precision. Rest layers are fused into Convolutions using post operations optimization technique, which is described in [Internal CPU Plugin Optimizations](supported_plugins/CPU.md).
-
-[int8_flow]: img/cpu_int8_flow.png