We are using PaddlePaddle to train our models, but TensorRT cannot directly parse Paddle Models, so we use Paddle2ONNX to transfer Paddle models to ONNX models, then use TensorRT ONNX parser to parse our models indirectly
But we encounter one issue, some ONNX models cannot be parsed by TensorRT although most of them can be parsed, the issue is as below:
[libprotobuf ERROR google/protobuf/text_format.cc:298] Error parsing text-format onnx2trt_onnx.ModelProto: 1:1: Invalid control characters encountered in text.
[libprotobuf ERROR google/protobuf/text_format.cc:298] Error parsing text-format onnx2trt_onnx.ModelProto: 1:17: Message type “onnx2trt_onnx.ModelProto” has no field named “PaddlePaddle”.
Failed to parse ONNX model from filetreatment_classification.onnx
Network must have at least one output
could not build engine
trtexec: trtexec.cpp:338: nvinfer1::ICudaEngine* onnxToTRTModel(): Assertion `false’ failed.
This issue has blocked us moving forward, please help to check it