Skip to content

🐛 [Bug] torch._export.verifier.SpecViolationError when using Torch-TensorRT #2655

Open
@DefTruth

Description

@DefTruth

Bug Description

got the export error with torch_tensorrt 2.2.0 use dynamo IR

torch._export.verifier.SpecViolationError: Node.meta max_pool2d_default is missing val field.

To Reproduce

optimized_model = torch_tensorrt.compile(
                        self.model,
                        ir="dynamo",
                        inputs=inputs,
                        enabled_precisions=enabled_precisions,
                        debug=debug,
                        workspace_size=(2147483648) * 4, # 8GB,
                        min_block_size=7,
                        torch_executed_ops=torch_executed_ops,
                        truncate_long_and_double=True,
                        enable_experimental_decompositions=True
 )

Steps to reproduce the behavior:

Expected behavior

Environment

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

  • Torch-TensorRT Version 2.2.0:
  • PyTorch Version 2.2.1:
  • CPU Architecture: x86_64
  • OS (e.g., Linux): Linux
  • How you installed PyTorch (conda, pip, libtorch, source): pip
  • Build command you used (if compiling from source): no
  • Are you using local sources or building from archives: no
  • Python version: 3.10
  • CUDA version: cuda 12.3
  • GPU models and configuration: A30
  • Any other relevant information: no

Additional context

Metadata

Metadata

Assignees

Labels

bugSomething isn't working

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions