It would be beneficial to write a design document on INT8 DNNL implementation and put it into
FluidDoc next to other Design documents:
https://github.com/PaddlePaddle/FluidDoc/tree/develop/doc/fluid/design/mkldnn

• 系统环境：

  • Paddle版本：1.5.1，CPU，无使用其他加速模块
  • 系统: CentOS 6.3

• 问题描述：

  • 使用paddle.fluid.contrib.slim.Compressor模块进行模型压缩
  • 压缩后的模型，float能正常运行，int8版出现以下错误：
    

• 问题复现：

git clone https://github.com/Bond-SYSU/paddle_compress.git
cd paddle_compress
sh run.sh compress   #

Changes made in #1874 to test/models/mxnet/1rnn_layer_3lstm.json reveal an error in the fusion of a sequence of LSTM cells into a single layer RNN for gpu.

The fused result should look like:

Instead, we currently see:
![rnn_fusion_issue](https://user-images.githubusercontent.com/48

The "paddle.reader.Fake" class can be found in English document, in the following link:
https://www.paddlepaddle.org.cn/documentation/docs/en/1.5/api/data/data_reader.html
It is just below "multiprocess" class, with demo codes:

But in the Chinese version:
https://www.paddlepaddle.org.cn