ExampleAsyncMLInferenceWithTriton.h

Go to the documentation of this file.

// Copyright (C) 2002-2026 CERN for the benefit of the ATLAS collaboration
 
#pragma once
 
// Framework include(s).
#include "AthenaBaseComps/AthAsynchronousAlgorithm.h"
#include "AthOnnxInterfaces/IAthInferenceTool.h"
 
// System include(s).
#include <string>
#include <vector>
 
namespace AthInfer {

   class ExampleAsyncMLInferenceWithTriton: public AthAsynchronousAlgorithm {
 
   public:
      using AthAsynchronousAlgorithm::AthAsynchronousAlgorithm;


      virtual StatusCode initialize() override;
      virtual StatusCode execute( const EventContext& ctx ) const override;

 
   private:

      Gaudi::Property< std::string > m_pixelFileName{ this, "InputDataPixel",
         "dev/MLTest/2020-03-31/t10k-images-idx3-ubyte",
         "Name of the input pixel file to load" };

      Gaudi::Property<int> m_batchSize {this, "BatchSize", 1, "No. of elements/example in a batch"};

      ToolHandle< AthInfer::IAthInferenceTool >  m_tritonTool{
         this, "InferenceTool", "AthInfer::TritonTool", "Triton client tool"
      };
 
      std::vector<std::vector<std::vector<float>>> m_input_tensor_values_notFlat;
 
   private:
      std::vector<std::vector<std::vector<float>>> read_mnist_pixel_notFlat(const std::string &full_path) const;
      std::vector<float> flattenNestedVectors(const std::vector<std::vector<float>>& nestedVector) const;
   };
 
} // namespace AthInfer