Wanted to know in detail about parser in the tensorrt

[trivedi.nagaraj]

Hi all, this is regarding the minimal information present in the tensorrt about how the parser function works.
While understanding the trtexec source and in many of the source code I have come across with these functions and have got following doutbs

1. parser.caffeParser->parse(
   model.prototxt.c_str(), modelName.empty() ? nullptr : modelName.c_str(), network, DataType::kFLOAT);

2. parser.uffParser->parse(model.baseModel.model.c_str(), network);

3. parser.onnxParser.reset(createParser(network, sample::gLogger.getTRTLogger()));
   parser.onnxParser->parseFromFile(
   model.baseModel.model.c_str(), static_cast(sample::gLogger.getReportableSeverity())

Doubt - What it does with the parameter ‘network’ passed to it?. Will it parse the model in terms of understanding the layers present in the model and populate the network [creates layers in the network and assigns number of input/output and weights to each of the layers?

---

Please clarify my above doubts.

Also the tensorrt documentation I have read on the parser, I am not finding a detailed explanation on parser. Can you please point me to a right place in the documentation where I find a detailed information on parser with regards to parser.

Thanks and Regards

Nagaraj Trivedi

[trivedi.nagaraj]

Hi all, please update me on this.

Thanks and Regards

Nagaraj Trivedi

[SivaRamaKrishnaNV]

Dear @trivedi.nagaraj,
We have open sourced the parsers at https://github.com/NVIDIA/TensorRT/tree/release/8.6/parsers . You can check the implementation details.