Test inference performance on GPU,

[YHRen]

@pphuangyi

• [x] torch scripts to save and load trained-model.
• [x] model load and run inference.

After this is done, I will contact GraphCore to have a walk through of the code.

[pphuangyi]

• [x] torch scripts to save and load trained-model.
  • x in the utils folder: script trained model;
  • [x] [TPC data torch scriptor] in the utils folder: script the input data.
• [ ] model load and run inference.
  • [x] Load scripted encoder and run inference in python;
  • [x] Load scripted encoder and run inference in C;
  • [ ] Check whether they give the same result;
  • [ ] Compare inference speed.
• [ ] TorchScript with GPU;
• [ ] Using using half precision for inference.