❓ [Question] Can you convert model that operates on custom classes?

[MarekPokropinski]

❓ Question

I have a torch module that creates objects of custom classes that have tensors as fields. It can be torch.jit.scripted but torch.jit.trace can be problematic. When I torch.jit.script module and then torch_tensorrt.compile it I get the following error: Unable to get schema for Node %317 : __torch__.src.MyClass = prim::CreateObject() (conversion.VerifyCoverterSupportForBlock)

What you have already tried

torch.jit.trace avoids the problem but introduces problems with loops in module.

Environment

Build information about Torch-TensorRT can be found by turning on debug messages

• PyTorch Version (e.g., 1.0): 1.10.2
• CPU Architecture: intel
• OS (e.g., Linux): linux
• How you installed PyTorch (conda, pip, libtorch, source): pip
• Build command you used (if compiling from source):
• Are you using local sources or building from archives: from archives
• Python version: 3.8
• CUDA version: 11.3
• GPU models and configuration: rtx 3090
• Any other relevant information:

[narendasan]

You should be able to use the method in this issue to compile your model #876

[narendasan]

Also make sure your operation is registered through torchbind https://pytorch.org/tutorials/advanced/torch_script_custom_ops.html