模型部署 修改

This commit is contained in:
limin 2025-02-13 14:01:23 +08:00
parent 5b089fb404
commit 53e7f47065
5 changed files with 14 additions and 36 deletions

View File

@ -13,7 +13,8 @@ public enum ModelDeployConstantEnum implements IntArrayValuable {
MODEL_DEPLOY_PENDING(1,"pending"),
MODEL_DEPLOY_RUNNING(2,"running"),
MODEL_DEPLOY_SUCCESS(2,"success"),
MODEL_DEPLOY_FAILED(3,"failed");
MODEL_DEPLOY_FAILED(3,"failed"),
MODEL_DEPLOY_STOP(1,"stop");
public static final int[] ARRAYS = Arrays.stream(values()).mapToInt(ModelDeployConstantEnum::getStatus).toArray();
/**

View File

@ -77,30 +77,10 @@ public class AsyncModelServiceService {
@Async
public void startModelService(ModelServiceDO updateObj) {
try {
/**
* cpu:1,
* gpu:0
* inferredType:"cpu"
* label:"cpu-aigc-model"
* maxGpuMemory:0
* modelWorker:""
* quantization:"float16"
* replicas:1
*/
/**
* {
* "model": "string",
* "inferredType": "gpu",
* "cpu": 0,
* "gpu": 1,
* "replicas": 1,
* "label": "cpu-aigc-model",
* "quantization": "float16",
* "maxGpuMemory": 0,
* "modelWorker": ""
* }
*/
ModelServiceDO modelServiceDO = modelServiceMapper.selectById(updateObj.getId());
if (modelServiceDO.getJobId() != null){
AigcModelDeployVO modelDeployVO = trainHttpService.modelUndeploy(new HashMap<>(), modelServiceDO.getJobId());
}
AigcModelDeploySaveReq aigcModelDeploySaveReq = new AigcModelDeploySaveReq(modelServiceDO.getBaseModelName(),
"gpu");
ModelDeployRespVO modelDeployRespVO = trainHttpService.modelDeploy(new HashMap<>(), aigcModelDeploySaveReq);
@ -120,11 +100,9 @@ public class AsyncModelServiceService {
try {
ModelServiceDO modelServiceDO = modelServiceMapper.selectById(updateObj.getId());
// Thread.sleep(30000);
AigcRespVO aigcRespVO = trainHttpService.modelUndeploy(new HashMap<>(), modelServiceDO.getJobId());
if (aigcRespVO.isSuccess()){
AigcModelDeployVO modelDeployRespVO = trainHttpService.modelUndeploy(new HashMap<>(), modelServiceDO.getJobId());
if (modelDeployRespVO.getStatus().equals("stop")){
updateObj.setStatus(0);
}else {
// updateObj.setStatus(3);
}
modelServiceMapper.updateById(updateObj);
}catch(Exception e){

View File

@ -246,14 +246,14 @@ public class TrainHttpService {
ModelDeployRespVO modelDeployRespVO = JSON.parseObject(res.getBytes(), ModelDeployRespVO.class);
return modelDeployRespVO;
}
public AigcRespVO modelUndeploy(Map<String, String> headers, Long deployId) {
public AigcModelDeployVO modelUndeploy(Map<String, String> headers, Long deployId) {
login(headers);
String modelDeploy = llmBackendProperties.getModelUndeploy();
String res = HttpUtils.post(modelDeploy + deployId, headers,"");
log.info(" modelDeploy:{}", res);
AigcRespVO aigcRespVO = JSON.parseObject(res, AigcRespVO.class);
log.info(" modelDeploy:{}", aigcRespVO);
return aigcRespVO;
AigcModelDeployVO modelDeployRespVO = JSON.parseObject(res, AigcModelDeployVO.class);
log.info(" modelDeploy:{}", modelDeployRespVO);
return modelDeployRespVO;
}
/**

View File

@ -37,12 +37,12 @@ public class AigcModelDeployVO {
}*/
@JsonProperty("updated_at")
// @JsonDeserialize(using = AigcCustomDateTimeDeserializer.class)
private LocalDateTime updatedAt;
private String updatedAt;
@JsonProperty("model_id")
private Integer modeId;
@JsonProperty("created_at")
// @JsonDeserialize(using = AigcCustomDateTimeDeserializer.class)
private LocalDateTime createdAt;
private String createdAt;
private String status;
@JsonProperty("inferred_type")
private String inferredType;
@ -52,7 +52,7 @@ public class AigcModelDeployVO {
private String deploymentName;
@JsonProperty("deleted_at")
// @JsonDeserialize(using = AigcCustomDateTimeDeserializer.class)
private LocalDateTime deletedAt;
private String deletedAt;
private Integer id;
@JsonProperty("model_path")
private String modelPath;

View File

@ -71,7 +71,6 @@ public class ModelServiceTaskSyncService {
module.addDeserializer(LocalDateTime.class, new AigcCustomDateTimeDeserializer());
mapper.registerModule(module);
List<AigcModelDeployVO> aigcModelDeployVOS = mapper.readValue(res,new TypeReference<List<AigcModelDeployVO>>() {});
Collections.sort(aigcModelDeployVOS, Comparator.comparing(AigcModelDeployVO::getUpdatedAt).reversed());
// 获取最新的一条记录
if (!aigcModelDeployVOS.isEmpty()) {
AigcModelDeployVO latestRecord = aigcModelDeployVOS.get(0);