tensorflow/tensorrt

Internal error: could not find any implementation for node conv2d_transpose_1, try increasing the workspace size with IBuilder::setMaxWorkspaceSize()

diligent321 opened this issue · 2 comments

diligent321 commented

Hello,
when i convert my pb model to tftrt FP16, it raised the following error,
"2019-09-03 12:23:15.537205: E tensorflow/compiler/tf2tensorrt/utils/trt_logger.cc:41] DefaultLogger Internal error: could not find any implementation for node generator/generator_unit/conv_tran2highres/conv_tran1/Conv2d_transpose/conv2d_transpose_1, try increasing the workspace size with IBuilder::setMaxWorkspaceSize()".

then, i checked my pb model, as the following image show, when i set the output_name before Conv2DBackpropInput, it generate the TRTEngineOp sucessfully. However, when i set the output_name after Conv2DBackpropInput, it failed. so what's the problem, i'am really appreciated for your kindly explanation.

For the environment,
Hardware: T4 card
Tensorflow: nvidia-tensorflow-19.06-py3 (docker image)

捕获

NurzatRakhman commented

The same error occurred to me with conv2d_transpose layer.

  • 👍1
pooyadavoodi commented

Please update TF to >=1.14 (NVIDIA container >=19.07).
That version of TF supports conv2d_transpose: https://docs.nvidia.com/deeplearning/frameworks/tf-trt-user-guide/index.html#supported-ops

Closing. Please reopen if you still see the issue.