SaltModelTriton.cxx File Reference

#include "FlavorTagInference/SaltModelTriton.h"
#include "FlavorTagInference/SaltModelGraphConfig.h"
#include "FlavorTagInference/SaltModelOutput.h"
#include "CxxUtils/checker_macros.h"
#include <onnxruntime_cxx_api.h>
#include <stdexcept>
#include <tuple>
#include <set>
#include <memory>

Include dependency graph for SaltModelTriton.cxx:

Go to the source code of this file.

Classes
struct	TritonDType< float >
struct	TritonDType< int64_t >

Namespaces
namespace	FlavorTagInference
	This file contains "getter" functions used for accessing tagger inputs from the EDM.

Functions
template<typename T>
bool	prepareInput (const std::string &name, const std::vector< int64_t > &shape, const std::vector< T > &data, std::vector< std::shared_ptr< tc::InferInput > > &inputs)
template<typename T>
bool	extractOutput (const std::string &name, const std::shared_ptr< tc::InferResult > &result, std::vector< T > &outputVec)

Function Documentation

extractOutput()

template<typename T>

bool extractOutput	(	const std::string &	name,
		const std::shared_ptr< tc::InferResult > &	result,
		std::vector< T > &	outputVec )

Definition at line 55 of file SaltModelTriton.cxx.

{
  const uint8_t* rawData = nullptr;
  size_t size = 0;
 
  // Get access to the buffer holding raw results of specified output returned by the server.
  // Note: the buffer is owned by InferResult instance.
  // Users can copy out the data if required to extend the lifetime.
  tc::Error err = result->RawData(name, &rawData, &size);
  if(!err.IsOk()) {
    std::cerr << "Unable to get raw output for: " + name << std::endl;
    return false;
  }
 
  outputVec.resize(size / sizeof(T));
  std::memcpy(outputVec.data(), rawData, size);
  return true;
}

prepareInput()

template<typename T>

bool prepareInput	(	const std::string &	name,
		const std::vector< int64_t > &	shape,
		const std::vector< T > &	data,
		std::vector< std::shared_ptr< tc::InferInput > > &	inputs )

Definition at line 22 of file SaltModelTriton.cxx.

{
  const char* dtype = TritonDType<T>::value;
  tc::InferInput* rawInputPtr = nullptr;
 
  // create the InferInput object with the predefined name, shape, and data type.
  tc::Error err = tc::InferInput::Create(&rawInputPtr, name, shape, dtype);
  if(!err.IsOk()) {
    std::cerr << "Unable to create input: " + name << std::endl;
    return false;
  }
 
  // Append tensor values for this input from a byte array.
  // Note: The vector is not copied and so it must not be modified or destroyed
  // until this input is no longer needed (that is until the Infer() call(s) that use the input have completed).
  // Multiple calls can be made to this API to keep adding tensor data for this input.
  // The data will be delivered in the order it was added.
  std::shared_ptr<tc::InferInput> input(rawInputPtr);
  err = input->AppendRaw(reinterpret_cast<const uint8_t*>(data.data()),
             data.size() * sizeof(T));
  if(!err.IsOk()) {
    std::cerr << "Unable to set input data for: " + name << std::endl;
    return false;
  }
 
  inputs.push_back(std::move(input));
  return true;
}