[DNNL] Evaluate trained paddle models using mkl-dnn kernels
Created by: jczaja
This task is to check trained model of PaddlePaddle: https://github.com/PaddlePaddle/models in order to locate possible room for improvements of our integration.
Scope:
- Check if possible to run inference on CPU engine. a) write down performance if available
- Check if possible to run inference on MKL-DNN engine. a) Note , problems if any b) note performance if available
- Get graph (dot, pdf)
- Check if possible to train that model
- Note which dataset available