Intermediate Representation Suitable for INT8 Inference

## Introduction

Inference Engine CPU plugin can infer models in the 8-bit integer (INT8) precision. For details, refer to INT8 inference on the CPU.

Intermediate Representation (IR) should be specifically formed to be suitable for the INT8 inference. Such an IR is called an INT8 IR and you can generate it in two ways:

For an operation to be executed in INT8, it must have FakeQuantize operations as inputs with the levels attribute set to 255 or 256. See the specification of FakeQuantize operation for details. To see the list of supported INT8 layers, refer to INT8 inference on the CPU.

To execute the Convolution operation in INT8 on CPU, both data and weight inputs should have FakeQuantize as an input operation:

INT8 IR is also suitable for FP32 and FP16 inference if a chosen plugin supports all operations of the IR, because the only difference between an INT8 IR and FP16 or FP32 IR is the existence of FakeQuantize in the INT8 IR. Plugins with the INT8 inference support recognize these sub-graphs and quantize them during the inference time. Plugins without the INT8 support execute all operations, including FakeQuantize, as is in the FP32 or FP16 precision.

Accordingly, the presence of FakeQuantize operations in the IR is a recommendation for a plugin on how to quantize particular operations in the model. If capable, a plugin accepts the recommendation and performs the INT8 inference, otherwise the plugin ignores the recommendation and executes a model in the floating-point precision.

## Compressed INT8 Weights

Weighted operations, like Convolution, MatMul, and others, store weights as floating-point Constant in the graph followed by the FakeQuantize operation. Constant followed by the FakeQuantize operation could be optimized memory-wise due to the FakeQuantize operation semantics. The resulting weights sub-graph stores weights in INT8 Constant, which gets unpacked back to floating point with the Convert operation. Weights compression leaves FakeQuantize output arithmetically the same and weights storing takes four times less memory.

See the visualization of Convolution with the compressed weights:

Both Model Optimizer and Post-Training Optimization tool generate a compressed IR by default. To generate an expanded INT8 IR, use --disable_weights_compression.