Quasar

GPT3-enabled surrogate functions in modern C++

Installation

• Dependencies: libcurl, nlohmann_json 3.2.0
• Install process: cmake .; make; make install

Example usage

#include <quasar.h>
using namespace quasar; 

...

// Load OpenAI API key from file.
std::ifstream fs("<api_key_file>.json");
std::string key;
if(fs.good())
    getline(fs, key);

// Must initialize before using
if(Quasar::init(key)) return 1;

std::string bad = "What day of the wek is it?";
std::string good;

Quasar("Fix the spelling mistake in bad and output it into good.")
    .with_model(Model::GPT_Davinci) // Optional params: gpt temperature, max tokens
    .bind_input<std::string>("bad", &bad) // Input can be any streamable type
    .execute()
    .extract<std::string>("good", &good); // Extraction throws exception on failure

std::cout << good; // "What day of the week is it?"

Quasar::cleanup();

Notes

• GPT3 is not necessarily deterministic. Visit the OpenAI docs to learn more about temperature and tokens.
• Quasar inputs and outputs should ideally be wrapped in try/catch blocks.