When working with python, I am not finding some of the calls and functionality I see available in the C++ API.
For example, under “/usr/tensorrt/samples/common/sampleEngines.cpp” I see
TrtUniquePtr runtime{createInferRuntime(gLogger.getTRTLogger())};
runtime->setDLACore(DLACore);
In the python runtime tensorrt.Runtime
I dont see any calls to setDLACore. Only in IBuilderConfig do I see how to put a model on a DLA, but correct me if I am wrong, isnt this class for building a TensorRT model from another framework, like onnx? How would I load an existing TensortRT model and run it on the DLA like the above C++ code does?
Similarly, in the C++ api I see IBuilderConfig::setMaxWorkspaceSize()
The following builder.max_workspace_size in python seems to have no affect
trt.Builder(TRT_LOGGER) as builder
builder.max_workspace_size = 1 << 25 #This doesnt seem to do anything
builder_cfg = builder.create_builder_config()
engine = builder.build_engine(network, builder_cfg)
To elaborate, building with builder.build_cuda_engine(network) works fine but using the builder config results in [TensorRT] ERROR: Try increasing the workspace …
Thank you.