ModularML

Lightweight and modular library for building cpp integrated machine learning models from onnx models.

Installation

Install the library and link it to your project using CMake. This is done by pasting the following code into your CMakeLists.txt file.

# ----------------------- MODULARML -------------------------- #
include(FetchContent)
FetchContent_Declare(
    modularml
    GIT_REPOSITORY https://github.com/willayy/modularml-lib
    GIT_TAG        <use latest release tag>
)
FetchContent_MakeAvailable(modularml)
target_link_libraries(MyProject PRIVATE modularml)
# ------------------------------------------------------------ #

Usage

#include <modularml>
#include <memory>
#include <iostream>

// Define the blocked gemm flag to enable the compilation of the blocked gemm function
#define USE_BLOCKED_GEMM

int main() {

  // Dynamically change the gemm function pointer to point to the blocked version, this works for any function 
  // with a matching signature.
  TensorOperations::set_gemm_ptr<int, float, double>(
    mml_gemm_blocked<int>, 
    mml_gemm_blocked<float>, 
    mml_gemm_blocked<double>
  );

  // Use the TensorFactory to create tensors
  auto a = TensorFactory::create_tensor<int>({2, 3}, {1, 2, 3, 4, 5, 6});
  auto b = TensorFactory::create_tensor<int>({3, 2}, {7, 8, 9, 10, 11, 12});
  auto c = TensorFactory::create_tensor<int>({2, 3});

  // Perform the gemm operation using the blocked version.
  TensorOperations::gemm<int>(0, 0, 2, 2, 3, 1, a, 3, b, 2, 1, c, 2);

  // Print the result, ensuring that the computation works.
  std::cout << "Result of gemm: " << (*c) << std::endl;

  return 0;
}

Usage with AlexNet

#include <modularml>
#include <memory>
#include <iostream>
#include <fstream>

int main() {
  /* Load the json file that has been created by onnx2json, that script can be downloaded externally
  Or you can use the one thats found in build/_deps/modularml-src/onnx2json */
  std::ifstream file("../src/alexnet_trained.json");
  nlohmann::json json_model;
  file >> json_model;
  file.close();
  // Parse the model into a model object, containing all the nodes.
  auto parser = Parser_mml();
  auto model = parser.parse(json_model);
  // Create an input tensor thats empty
  auto input = TensorFactory::create_tensor<float>({1, 3, 224, 224});
  // Create an input map
  std::unordered_map<std::string, GeneralDataTypes> input_map;
  input_map["input"] = input;
  // Run inference with the input map
  auto output_map = model->infer(input_map);
  // Get the output
  auto output = std::get_if<std::shared_ptr<Tensor<float>>>(&(output_map["output"]));
  // Extract the output tensor and print it!
  std::cout << "Output: " << (*output) << std::endl;
  return 0;
}

Contributing

We welcome contributions!
Please read our Contributing Guide for instructions on how to get started.

License

This project is licensed under the MIT License.

Acknowledgments

This library is a fork from the framework that was developed as part of a Bachelor Thesis at Chalmers University of Technology. Which can be found here