Latency improvement with Ipex

[kiko0217]

in other case, Doctr use in server without GPU so must improve speed. maybe. can Doctr be approved with Ipex?, or make version lite.
https://github.com/intel/intel-extension-for-pytorch

[felixdittrich92]

Hi @kiko0217,
we have on track for 0.6.0 to make all models ONNX exportable and providing a way to use this exported models directly in docTR 👍 (than your "lite" models will be the quantized versions)