* Copyright (c) Huawei Technologies Co., Ltd. 2025. All rights reserved.
* MindIE is licensed under Mulan PSL v2.
* You can use this software according to the terms and conditions of the Mulan PSL v2.
* You may obtain a copy of Mulan PSL v2 at:
* http://license.coscl.org.cn/MulanPSL2
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
* EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
* MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
* See the Mulan PSL v2 for more details.
*/
#pragma once
#include <thread>
#include "base_config_manager.h"
#include "cmath"
#include "common_util.h"
#include "config_interaction.h"
#include "config_manager.h"
#include "env_util.h"
namespace mindie_llm {
const ServerConfig &GetServerConfig();
const std::vector<ModelDeployConfig> &GetModelDeployConfig();
const BackendConfig &GetBackendConfig();
const ScheduleConfig &GetScheduleConfig();
const std::vector<LoraConfig> &GetLoraConfig();
const RanktableParam &GetRanktableParam();
class ConfigManager::Impl {
public:
Impl() = delete;
explicit Impl(const std::string &jsonPath);
Impl(const Impl &) = delete;
Impl operator=(const Impl &) = delete;
~Impl() {
running_ = false;
if (task_.joinable()) {
task_.join();
}
}
[[nodiscard]] const BackendConfig &GetBackendConfig() const {
return std::atomic_load(&backendConfig_)->GetParam();
}
[[nodiscard]] const LogConfig &GetLogConfig() const { return std::atomic_load(&logConfig_)->GetParam(); }
[[nodiscard]] const ServerConfig &GetServerConfig() const { return std::atomic_load(&serverConfig_)->GetParam(); }
[[nodiscard]] const std::vector<ModelDeployConfig> &GetModelDeployConfig() const {
return std::atomic_load(&modelDeployConfig_)->GetParam();
}
[[nodiscard]] const std::vector<LoraConfig> &GetLoraConfig() const {
return std::atomic_load(&modelDeployConfig_)->GetLoraConfig();
}
[[nodiscard]] const ScheduleConfig &GetScheduleConfig() const {
return std::atomic_load(&scheduleConfig_)->GetParam();
}
[[nodiscard]] const RanktableParam &GetRanktableParam() const {
return std::atomic_load(&ranktableConfig_)->GetParam();
}
[[nodiscard]] bool CheckAllParam();
bool CheckAndInitLogParam();
static bool CreateInstance(std::string jsonPath = "");
static ConfigManager::Impl &GetInstance();
void SetBlockNum(uint32_t cpuBlockNum, uint32_t npuBlockNum) {
scheduleConfig_->SetBlockNum(cpuBlockNum, npuBlockNum);
}
void SetMaxPositionEmbeddings(uint32_t maxPositionEmbeddings) {
if (maxPositionEmbeddings > modelDeployConfig_->GetParam()[0].maxPositionEmbeddings) {
modelDeployConfig_->SetMaxPositionEmbeddings(maxPositionEmbeddings);
}
}
void SetTokenTimeout(uint64_t tokenTimeout) { serverConfig_->SetTokenTimeout(tokenTimeout); }
void SetE2eTimeout(uint64_t e2eTimeout) { serverConfig_->SetE2eTimeout(e2eTimeout); }
std::string GetConfigJsonStr();
bool IsMultiNodeInfer() const;
bool IslayerwiseDisaggregated() const;
bool IsLwdMultiNodesEnable() const;
std::string GetLwdRoleType() const;
void UpdateConfig();
void ExecuteConfigInteractions();
private:
[[nodiscard]] bool PreCheck() const;
[[nodiscard]] bool CheckLayerwiseDisaggregatedParam();
bool InitConfigManager();
bool initialized_ = false;
std::shared_ptr<ScheduleConfigManager> scheduleConfig_{nullptr};
std::shared_ptr<LogConfigManager> logConfig_{nullptr};
std::shared_ptr<ServerConfigManager> serverConfig_{nullptr};
std::shared_ptr<BackendConfigManager> backendConfig_{nullptr};
std::shared_ptr<ModelDeployConfigManager> modelDeployConfig_{nullptr};
std::shared_ptr<RanktableConfigManager> ranktableConfig_;
bool hasChecked_{false};
std::string jsonPath_;
nlohmann::json configJson_{};
std::atomic<bool> running_{true};
std::thread task_;
};
}