humanus.cpp/llm.h

#ifndef HUMANUS_LLM_H
#define HUMANUS_LLM_H

#include "config.h"
#include "logger.h"
#include "schema.h"
#include "mcp/common/httplib.h"
#include <map>
#include <string>
#include <memory>
#include <vector>
#include <functional>
#include <stdexcept>
#include <future>

namespace humanus {

class LLM {
private:
    static std::map<std::string, std::shared_ptr<LLM>> _instances;

    std::unique_ptr<httplib::Client> client_;

    LLMSettings llm_config_;
    
    // 私有构造函数，防止直接创建实例
    LLM(const std::string& config_name, const LLMSettings llm_config) : llm_config_(llm_config) {
        client_ = std::make_unique<httplib::Client>(llm_config.base_url);
        client_->set_default_headers({
            {"Authorization", "Bearer " + llm_config_.api_key}
        });
    }

public:
    // 单例模式获取实例
    static std::shared_ptr<LLM> get_instance(const std::string& config_name = "default", const LLMSettings llm_config = LLMSettings()) {
        if (_instances.find(config_name) == _instances.end()) {
            _instances[config_name] = std::make_shared<LLM>(config_name, llm_config);
        }
        return _instances[config_name];
    }
    
    /**
     * @brief 格式化消息列表为LLM可接受的格式
     * @param messages Message对象消息列表
     * @return 格式化后的消息列表
     * @throws std::invalid_argument 如果消息格式无效或缺少必要字段
     * @throws std::runtime_error 如果消息类型不支持
     */
    static std::vector<json> format_messages(const std::vector<Message>& messages) {
        std::vector<json> formatted_messages;

        for (const auto& message : messages) {
            formatted_messages.push_back(message.to_json());
        }

        for (const auto& message : formatted_messages) {
            if (message["role"] != "user" && message["role"] != "assistant" && message["role"] != "system" && message["role"] != "tool") {
                throw std::invalid_argument("Invalid role: " + message["role"]);
            }
            if (message["content"].empty() && message["tool_calls"].empty()) {
                throw std::invalid_argument("Message must contain either 'content' or 'tool_calls'");
            }
        }

        return formatted_messages;
    }

    /**
     * @brief 格式化消息列表为LLM可接受的格式
     * @param messages json对象消息列表
     * @return 格式化后的消息列表
     * @throws std::invalid_argument 如果消息格式无效或缺少必要字段
     * @throws std::runtime_error 如果消息类型不支持
     */
    static std::vector<json> format_messages(const std::vector<json>& messages) {
        std::vector<json> formatted_messages;

        for (const auto& message : messages) {
            if (!message.contains("role")) {
                throw std::invalid_argument("消息缺少必要字段: role");
            }
            formatted_messages.push_back(message);
        }

        for (const auto& message : formatted_messages) {
            if (message["role"] != "user" && message["role"] != "assistant" && message["role"] != "system" && message["role"] != "tool") {
                throw std::invalid_argument("Invalid role: " + message["role"]);
            }
            if (message["content"].empty() && message["tool_calls"].empty()) {
                throw std::invalid_argument("Message must contain either 'content' or 'tool_calls'");
            }
        }

        return formatted_messages;
    }
    
    /**
     * @brief 向LLM发送请求并获取回复
     * @param messages 对话消息列表
     * @param system_msgs 可选的系统消息
     * @param max_retries 最大重试次数
     * @return 生成的assitant content
     * @throws std::invalid_argument 如果消息无效或回复为空
     * @throws std::runtime_error 如果API调用失败
     */
    std::string ask(
        const std::vector<Message>& messages,
        const std::vector<Message>& system_msgs = {},
        int max_retries = 3
    ) {
        std::vector<json> formatted_messages;

        if (!system_msgs.empty()) {
            auto system_formatted_messages = format_messages(system_msgs);
            formatted_messages.insert(formatted_messages.end(), system_formatted_messages.begin(), system_formatted_messages.end());
        }
        
        auto _formatted_messages = format_messages(messages);
        formatted_messages.insert(formatted_messages.end(), _formatted_messages.begin(), _formatted_messages.end());

        json body = {
            {"model", llm_config_.model},
            {"messages", formatted_messages},
            {"temperature", llm_config_.temperature},
            {"max_tokens", llm_config_.max_tokens}
        };
        
        std::string body_str = body.dump();

        int retry = 0;

        while (retry <= max_retries) {
            // send request
            auto res = client_->Post(llm_config_.end_point, body_str, "application/json");

            if (!res) {
                logger->error("Failed to send request: " + httplib::to_string(res.error()));
            } else if (res->status == 200) {
                try {
                    json json_data = json::parse(res->body);
                    return json_data["choices"][0]["message"]["content"].get<std::string>();
                } catch (const std::exception & e) {
                    logger->error("Failed to parse response: " + std::string(e.what()));
                }
            } else {
                logger->error("Failed to send request: status=" + std::to_string(res->status) + ", body=" + res->body);
            }

            retry++;

            // wait for a while before retrying
            std::this_thread::sleep_for(std::chrono::milliseconds(500));

            logger->info("Retrying " + std::to_string(retry) + "/" + std::to_string(max_retries));
        }

        throw std::runtime_error("Failed to get response from LLM");
    }
    
    /**
     * @brief 使用工具功能向LLM发送请求
     * @param messages 对话消息列表
     * @param system_msgs 可选的系统消息
     * @param timeout 请求超时时间（秒）
     * @param tools 工具列表
     * @param tool_choice 工具选择策略
     * @param max_retries 最大重试次数
     * @return 生成的assistant message (content, tool_calls)
     * @throws std::invalid_argument 如果工具、工具选择或消息无效
     * @throws std::runtime_error 如果API调用失败
     */
    json ask_tool(
        const std::vector<Message>& messages,
        const std::vector<Message>& system_msgs = {},
        const std::vector<json> tools = {},
        const std::string& tool_choice = "auto",
        int timeout = 60,
        int max_retries = 3
    ) {
        if (tool_choice != "none" && tool_choice != "auto" && tool_choice != "required") {
            throw std::invalid_argument("Invalid tool_choice: " + tool_choice);
        }

        std::vector<json> formatted_messages;

        if (!system_msgs.empty()) {
            auto system_formatted_messages = format_messages(system_msgs);
            formatted_messages.insert(formatted_messages.end(), system_formatted_messages.begin(), system_formatted_messages.end());
        }
        
        auto _formatted_messages = format_messages(messages);
        formatted_messages.insert(formatted_messages.end(), _formatted_messages.begin(), _formatted_messages.end());

        if (!tools.empty()) {
            for (const json& tool : tools) {
                if (!tool.contains("type")) {
                    throw std::invalid_argument("Tool must contain 'type' field");
                }
            }
        }
        
        json body = {
            {"model", llm_config_.model},
            {"messages", formatted_messages},
            {"temperature", llm_config_.temperature},
            {"max_tokens", llm_config_.max_tokens},
            {"tools", tools},
            {"tool_choice", tool_choice}
        };

        client_->set_read_timeout(timeout);
        
        std::string body_str = body.dump();

        int retry = 0;

        while (retry <= max_retries) {
            // send request
            auto res = client_->Post(llm_config_.end_point, body_str, "application/json");

            if (!res) {
                logger->error("Failed to send request: " + httplib::to_string(res.error()));
            } else if (res->status == 200) {
                try {
                    json json_data = json::parse(res->body);
                    return json_data["choices"][0]["message"];
                } catch (const std::exception & e) {
                    logger->error("Failed to parse response: " + std::string(e.what()));
                }
            } else {
                logger->error("Failed to send request: status=" + std::to_string(res->status) + ", body=" + res->body);
            }

            retry++;

            // wait for a while before retrying
            std::this_thread::sleep_for(std::chrono::milliseconds(500));

            logger->info("Retrying " + std::to_string(retry) + "/" + std::to_string(max_retries));
        }

        throw std::runtime_error("Failed to get response from LLM");
    }
};

} // namespace humanus

#endif // HUMANUS_LLM_H