{

const std::string path = "/media/E/lichunyu/bert.cpp/data/test_tokenizer/tokenizer.json";

const std::string fname = "/media/E/lichunyu/bert.cpp/data/ggml-model/ggml_model.ggml";

auto start_cpp_total = std::chrono::high_resolution_clock::now();

bert_ctx *ctx = new bert_ctx();

bert_model_load_from_ggml(fname, ctx->model);

ctx->tokenizer.from_file(path);

ctx->buf_compute.resize(64 * 1024 * 1024);

auto start_cpp = std::chrono::high_resolution_clock::now();

std::string text = "测试一下tokenizer返回的结果是否和python一致";

int n_threads = 4;

int classification = bert_predict(ctx, text, n_threads);

auto end_cpp = std::chrono::high_resolution_clock::now();

auto duration_cpp = std::chrono::duration_cast<std::chrono::milliseconds>(end_cpp - start_cpp).count();

auto duration_cpp_total = std::chrono::duration_cast<std::chrono::milliseconds>(end_cpp - start_cpp_total).count();

std::cout << std::endl

<< std::endl

<< std::endl;

std::cout << "C++&Rust classification: " << classification << " and ";

std::cout << "C++&Rust without loading cost: " << duration_cpp << "ms." << std::endl;

auto start_python_total = std::chrono::high_resolution_clock::now();

std::system("/media/E/lichunyu/miniconda3/envs/bom/bin/python /media/E/lichunyu/bert.cpp/test/test_inference.py");

auto end_python = std::chrono::high_resolution_clock::now();

auto duration_python_total = std::chrono::duration_cast<std::chrono::milliseconds>(end_python - start_python_total).count();

std::cout << "C++&Rust with loading cost: " << duration_cpp_total << "ms." << std::endl;

std::cout << "Python with loading cost: " << duration_python_total << "ms." << std::endl;

return 0;