❓ [Question] Is there a way to visualize the TRT model?

[Scass0807]

❓ Question

I'm wondering if there is a way to get the TRT model after compilation and visualize it. I trying to compare a PTQ model to a QAT model. I know I might have to do some further optimization just trying to visualize the graphs and see what is going on . Currently using DenseNet169

What you have already tried

I can visualize an ONNX graph but unsure of TRT

Environment

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

• PyTorch Version (e.g., 1.0): 1.10.0
• CPU Architecture: x86 (Intel Skylake)
• OS (e.g., Linux): Ubuntu 20.04
• How you installed PyTorch (conda, pip, libtorch, source): pip
• Build command you used (if compiling from source): N/A
• Are you using local sources or building from archives:
• Python version: 3.8.10
• CUDA version: 11.4
• GPU models and configuration: Tesla T4
• Any other relevant information:

Additional context

[narendasan]

@andi4191 Might have more insight, but I think the trtexec tool might have like a layer dump but there isnt something like Tensorboard for TensorRT right now.

[github-actions[bot]]

This issue has not seen activity for 90 days, Remove stale label or comment or this will be closed in 10 days

[andi4191]

@Scass0807 : You can try trtexec tool from TensorRT with option --dumpLayerInfo --verbose to list the layer details in your TRTEngine.

For quick reference: https://github.com/NVIDIA/TensorRT/blob/main/samples/common/sampleOptions.cpp#L1518

[narendasan]

@Scass0807 There is a new tool from the TensorRT team that might help https://github.com/NVIDIA/TensorRT/tree/main/tools/experimental/trt-engine-explorer

[xylcbd]

how to visualize the trt model which is contained by the jit script checkpoint file?

[narendasan]

You can pull out the engine by referencing the attribute in the module similar to how youd access a submodule of a pytorch module