[TensorRT] ERROR: Internal error: could not find any implementation for node deconv1/conv2d_transpose, try increasing the workspace size with IBuilder::setMaxWorkspaceSize()
Workspace size defines the memory size that TensorRT library can use for constructing the model.
If size is too small, TensorRT will complain that none of its algorithms could fit into the given memory size.
You can find more information and tuning guide in our document:
5.2. Choosing the Optimal Workspace Size
thanks for your reply! I read these above as you said, but it did not give the method to fix my problem, what should i do, where is the function of “IBuilder::setMaxWorkspaceSize()”
Try to enlarge the workspace size:
For example, in sampleGoogleNet:
--- builder->setMaxWorkspaceSize(16 << 20); +++ builder->setMaxWorkspaceSize(16 << 24);