Zero Zeng

Could you please share a reproduce? Thanks!

I can also reproduce the issue on TRT 8.6 on x86. But in my test it's been fixed in TRT 10.

9.2 also pass.

Sorry for the late reply, I'm checking internally.

Could you please try mark `/model.0/conv/Conv ` or `/model.0/act/Sigmoid` as network output so that the layer fusion can be break? this can be done quickly with `polygraphy run model.onnx --mark...

I believe it's supported OOTB, you can export the model to onnx and check.

yes. even speed-up

OOTB should be supported. You can also use https://github.com/NVIDIA/TensorRT-LLM

@nvpohanh I'm kind of not sure about what the best practice for such issue after TRT-LLM released, could you please shed some light here :-D

Could you please try recompile the plugin with TRT 9.2? Thanks!