📖 [Story] Implement support for serde of TRT Graph modules with torch.export serde API

[peri044]

TL;DR

torch.export has serde API and now supports custom object (in our case torch.classes.tensorrt.Engine) serialization.

Goal(s)

We should be able to serialize/deserialize the output of torch-trt compilation.

Tasks

Tasks

Beta Give feedback

1. 🐛 [Bug] placeholder node meta data not preserved in fast partitioning #2258
   bug component: partitioning +2
   
2. 🐛 [Bug] Output node meta values is empty for TRT nodes [node_trt branch] #2259
   bug +1
   
3. 🐛 [Bug] Input bindings change b/w fast and global partitioner #2260
   bug +1
4. 🐛 [Bug] getitem nodes are expected as the output of TRT subgraph for torch.export serde to work #2261
   bug +1
5. ✨[Feature] Inline submodules so that torch.export serde can work #2270
   feature request +1
   
6. 🐛 [Bug] getitem node does not have meta data #2478
   bug +1
   
7. 🐛 [torch.export] VGG model export.load() failure #2479
   bug +1
8. 🐛 [Bug] Cannot save models using ExportedProgram if the model has weighted layers #2341
   Blocked [PyTorch] bug +2
   
9. PyTree based Collections I/O for ExportedPrograms #2553
   +0
   
10. 🐛 [Bug] Cannot perform inference if the ExportedProgram has weighted layers and custom ops. #2576
    bug +1
    

Additional context