Method Start

apps/mnncli/src/mnncli_server.cpp:224–491

}

void MnncliServer::Start(MNN::Transformer::Llm* llm, bool is_r1, const std::string& host, int port) {
    this->is_r1_ = is_r1;
    // Create a server instance
    httplib::Server server;

    // Define a route for the GET request on "/"
    server.Get("/", [this](const httplib::Request& req, httplib::Response& res) {
        AllowCors(res);
        res.set_content(html_content, "text/html");
    });
    server.Post("/reset", [&](const httplib::Request &req, httplib::Response &res) {
      LOG_DEBUG("POST /reset");
      AllowCors(res);
      llm->reset();
      res.set_content("{\"status\": \"ok\"}", "application/json");
    });
    
    server.Get("/v1/models", [&](const httplib::Request &req, httplib::Response &res) {
      LOG_DEBUG("GET /v1/models");
      AllowCors(res);
      json models_response = {
        {"object", "list"},
        {"data", json::array({
          {
            {"id", "ModelScope/MNN/Qwen2.5-0.5B-Instruct"},
            {"object", "model"},
            {"created", static_cast<int>(time(nullptr))},
            {"owned_by", "mnn"}
          }
        })}
      };
      res.set_content(models_response.dump(), "application/json");
    });
    server.Options("/v1/models", [](const httplib::Request& /*req*/, httplib::Response& res) {
        AllowCors(res);
        res.status = 200;
    });
    
    server.Options("/chat/completions", [](const httplib::Request& /*req*/, httplib::Response& res) {
        AllowCors(res);
        res.status = 200;
    });
    
    server.Options("/v1/chat/completions", [](const httplib::Request& /*req*/, httplib::Response& res) {
        AllowCors(res);
        res.status = 200;
    });
    // Handler function for chat completions
    auto chatCompletionsHandler = [&](const httplib::Request &req, httplib::Response &res) {
        LOG_DEBUG("POST chat/completions, handled by thread: " + std::to_string(std::hash<std::thread::id>{}(std::this_thread::get_id())));
      AllowCors(res);
      if (!json::accept(req.body)) {
          json err;
          err["error"] = "Invalid JSON in request body.";
          res.status = 400;
          res.set_content(err.dump(), "application/json");
          return;
      }