Merge branch 'master' of https://codeup.aliyun.com/63736f52e9565f4348a4cd42/xnjz-ai/xhllm
This commit is contained in:
commit
30be835063
@ -11,6 +11,7 @@ import cn.iocoder.yudao.module.llm.service.http.TrainHttpService;
|
||||
import cn.iocoder.yudao.module.llm.service.http.vo.AigcModelDeploySaveReq;
|
||||
import cn.iocoder.yudao.module.llm.service.http.vo.AigcModelDeployVO;
|
||||
import cn.iocoder.yudao.module.llm.service.http.vo.ModelDeployRespVO;
|
||||
import com.alibaba.fastjson.JSON;
|
||||
import com.alibaba.fastjson.JSONArray;
|
||||
import com.alibaba.fastjson.JSONObject;
|
||||
import com.esotericsoftware.minlog.Log;
|
||||
@ -19,6 +20,7 @@ import com.fasterxml.jackson.core.type.TypeReference;
|
||||
import com.fasterxml.jackson.databind.ObjectMapper;
|
||||
import com.fasterxml.jackson.databind.module.SimpleModule;
|
||||
import com.fasterxml.jackson.datatype.jsr310.JavaTimeModule;
|
||||
import lombok.extern.slf4j.Slf4j;
|
||||
import org.springframework.scheduling.annotation.Scheduled;
|
||||
import org.springframework.stereotype.Component;
|
||||
|
||||
@ -29,6 +31,7 @@ import java.util.List;
|
||||
import java.util.stream.Collectors;
|
||||
|
||||
@Component
|
||||
@Slf4j
|
||||
public class BaseModelTaskService {
|
||||
|
||||
@Resource
|
||||
@ -44,44 +47,90 @@ public class BaseModelTaskService {
|
||||
// 减少维护 先注释掉
|
||||
@Scheduled(cron ="0 0/1 * * * ?")
|
||||
public void synchronous() throws JsonProcessingException {
|
||||
List<BaseModelDO> baseModelList = baseModelService.getBaseModelList();
|
||||
for (BaseModelDO baseModelDO : baseModelList) {
|
||||
Long modelId = baseModelDO.getModelId();
|
||||
Long gpuId = baseModelDO.getGpuId();
|
||||
ServerNameDO serverNameDO1 = serverNameMapper.selectById(gpuId);
|
||||
String query = "?filter={\"id\":" + modelId + "}";
|
||||
String res = trainHttpService.modelTableQuery(new HashMap<>(), serverNameDO1.getHost(),"model_deploy", query);
|
||||
try {
|
||||
log.info("开始同步基础模型信息...");
|
||||
|
||||
ObjectMapper mapper = new ObjectMapper();
|
||||
mapper.registerModule(new JavaTimeModule());
|
||||
SimpleModule module = new SimpleModule();
|
||||
module.addDeserializer(LocalDateTime.class, new AigcCustomDateTimeDeserializer());
|
||||
mapper.registerModule(module);
|
||||
// 获取gpuhost主机
|
||||
ServerNameDO serverNameDO = serverNameMapper.selectById(baseModelDO.getGpuId());
|
||||
List<AigcModelDeployVO> aigcModelDeployVOS = mapper.readValue(res,new TypeReference<List<AigcModelDeployVO>>() {});
|
||||
if (!aigcModelDeployVOS.isEmpty()) {
|
||||
AigcModelDeployVO latestRecord = aigcModelDeployVOS.get(0);
|
||||
String status = latestRecord.getStatus();
|
||||
if(status.equals("stop")){
|
||||
// 获取所有基础模型列表
|
||||
log.debug("正在查询所有基础模型列表...");
|
||||
List<BaseModelDO> baseModelList = baseModelService.getBaseModelList();
|
||||
log.info("成功查询到 {} 个基础模型。", baseModelList.size());
|
||||
|
||||
AigcModelDeploySaveReq aigcModelDeploySaveReq = new AigcModelDeploySaveReq(baseModelDO.getAigcModelName(),
|
||||
"gpu");
|
||||
ModelDeployRespVO modelDeployRespVO = trainHttpService.modelDeploy(new HashMap<>(),serverNameDO.getHost(), aigcModelDeploySaveReq);
|
||||
if (!modelDeployRespVO.getMessage().equals("error")) {
|
||||
BaseModelSaveReqVO baseModelSaveReqVO = new BaseModelSaveReqVO();
|
||||
baseModelSaveReqVO.setId(baseModelDO.getId());
|
||||
baseModelSaveReqVO.setModelId(modelDeployRespVO.getId());
|
||||
baseModelSaveReqVO.setChatUrl(modelDeployRespVO.getPort() + DEFAULT_MODEL_URL_SUFFIX);
|
||||
baseModelService.updateBaseModel(new BaseModelSaveReqVO());
|
||||
// 遍历每个基础模型
|
||||
for (BaseModelDO baseModelDO : baseModelList) {
|
||||
Long modelId = baseModelDO.getModelId();
|
||||
Long gpuId = baseModelDO.getGpuId();
|
||||
log.debug("正在处理基础模型，模型ID: {}, GPU ID: {}", modelId, gpuId);
|
||||
|
||||
// 查询 GPU 服务器信息
|
||||
log.debug("正在查询 GPU 服务器信息，GPU ID: {}", gpuId);
|
||||
ServerNameDO serverName = serverNameMapper.selectById(gpuId);
|
||||
if (serverName == null) {
|
||||
log.error("未找到 GPU 服务器信息，GPU ID: {}", gpuId);
|
||||
continue;
|
||||
}
|
||||
log.debug("GPU 服务器信息查询成功。主机地址: {}", serverName.getHost());
|
||||
|
||||
// 构建查询参数并查询模型部署信息
|
||||
String query = "?filter={\"id\":" + modelId + "}";
|
||||
log.debug("正在查询模型部署信息，查询参数: {}", query);
|
||||
String res = trainHttpService.modelTableQuery(new HashMap<>(), serverName.getHost(), "model_deploy", query);
|
||||
log.debug("模型部署信息查询成功。响应内容: {}", res);
|
||||
|
||||
// 解析响应内容
|
||||
log.debug("正在解析模型部署信息...");
|
||||
ObjectMapper mapper = new ObjectMapper();
|
||||
mapper.registerModule(new JavaTimeModule());
|
||||
SimpleModule module = new SimpleModule();
|
||||
module.addDeserializer(LocalDateTime.class, new AigcCustomDateTimeDeserializer());
|
||||
mapper.registerModule(module);
|
||||
List<AigcModelDeployVO> aigcModelDeploys = mapper.readValue(res, new TypeReference<List<AigcModelDeployVO>>() {});
|
||||
log.debug("模型部署信息解析完成。记录数量: {}", aigcModelDeploys.size());
|
||||
|
||||
if (!aigcModelDeploys.isEmpty()) {
|
||||
AigcModelDeployVO latestRecord = aigcModelDeploys.get(0);
|
||||
String status = latestRecord.getStatus();
|
||||
log.debug("最新模型部署记录状态: {}", status);
|
||||
|
||||
// 如果模型状态为 "stop"，则重新部署
|
||||
if ("stop".equals(status)) {
|
||||
log.info("模型状态为 'stop'，正在重新部署模型...");
|
||||
|
||||
// 构建模型部署请求
|
||||
AigcModelDeploySaveReq aigcModelDeploySaveReq = new AigcModelDeploySaveReq(
|
||||
baseModelDO.getAigcModelName(), "gpu");
|
||||
log.debug("模型部署请求参数: {}", JSON.toJSONString(aigcModelDeploySaveReq));
|
||||
|
||||
// 发起模型部署请求
|
||||
ModelDeployRespVO modelDeployRespVO = trainHttpService.modelDeploy(
|
||||
new HashMap<>(), serverName.getHost(), aigcModelDeploySaveReq);
|
||||
log.debug("模型部署请求完成。响应内容: {}", JSON.toJSONString(modelDeployRespVO));
|
||||
|
||||
// 更新基础模型信息
|
||||
if (!"error".equals(modelDeployRespVO.getMessage())) {
|
||||
log.info("模型部署成功。正在更新基础模型信息...");
|
||||
BaseModelSaveReqVO baseModelSaveReqVO = new BaseModelSaveReqVO();
|
||||
baseModelSaveReqVO.setId(baseModelDO.getId());
|
||||
baseModelSaveReqVO.setModelId(modelDeployRespVO.getId());
|
||||
baseModelSaveReqVO.setChatUrl(modelDeployRespVO.getPort() + DEFAULT_MODEL_URL_SUFFIX);
|
||||
baseModelService.updateBaseModel(baseModelSaveReqVO);
|
||||
log.info("基础模型信息更新完成。模型ID: {}", baseModelDO.getId());
|
||||
} else {
|
||||
log.error("模型部署失败。模型ID: {}", baseModelDO.getId());
|
||||
}
|
||||
}
|
||||
} else {
|
||||
log.warn("未找到模型部署记录。模型ID: {}", modelId);
|
||||
}
|
||||
}
|
||||
|
||||
log.info("基础模型信息同步完成。");
|
||||
} catch (Exception e) {
|
||||
log.error("同步基础模型信息时发生异常。", e);
|
||||
throw e;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
||||
// @Scheduled(cron ="0 0/1 * * * ?")
|
||||
public void updateBaseModel() {
|
||||
Log.info("定时任务启动");
|
||||
|
||||
@ -263,6 +263,10 @@ public class FineTuningTaskServiceImpl implements FineTuningTaskService {
|
||||
.map(FineTuningTaskDO::getGpuType)
|
||||
.filter(Objects::nonNull)
|
||||
.collect(Collectors.toList());
|
||||
|
||||
if (CollectionUtils.isEmpty(gpuTypeIds)){
|
||||
return respVOS;
|
||||
}
|
||||
List<ServerNameDO> serverNameDOS = serverNameMapper.selectList(new LambdaQueryWrapper<ServerNameDO>()
|
||||
.in(ServerNameDO::getId, gpuTypeIds));
|
||||
Map<Long, ServerNameDO> longServerNameDOMap = cn.iocoder.yudao.framework.common.util.collection.CollectionUtils
|
||||
|
||||
@ -157,8 +157,8 @@ public class TrainHttpService {
|
||||
try {
|
||||
// 记录请求信息
|
||||
log.info("开始创建微调任务，请求URL: {}", url + llmBackendProperties.getFinetuningCreate());
|
||||
log.debug("请求头: {}", headers);
|
||||
log.debug("请求体: {}", JSON.toJSONString(req));
|
||||
log.info("请求头: {}", headers);
|
||||
log.info("请求体: {}", JSON.toJSONString(req));
|
||||
|
||||
// 发起 HTTP 请求
|
||||
log.debug("正在发起 HTTP POST 请求...");
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user