模型部署服务新增接口。

application-webadmin/src/main/java/com/yice/webadmin/app/constant/ModelConstant.java

@@ -10,4 +10,18 @@ public class ModelConstant {
      */
     public static final Integer DEPLOY_STATUS = 3;
 
+    /**
+     * 部署状态:成功
+     */
+    public static final Integer DEPLOY_STATUS_COMPLETE = 1;
+
+    /**
+     * 部署状态:部署中
+     */
+    public static final Integer DEPLOY_STATUS_DEPLOYED = 0;
+
+    /**
+     * 部署状态:失败
+     */
+    public static final Integer DEPLOY_STATUS_FAILURE = -1;
 }

application-webadmin/src/main/java/com/yice/webadmin/app/controller/ModelDeployController.java

@@ -166,7 +166,6 @@ public class ModelDeployController {
         ModelDeploy modelDeploy = MyModelUtil.copyTo(modelDeployDto, ModelDeploy.class);
         ModelDeploy originalModelDeploy = modelDeployService.getById(modelDeploy.getDeployId());
         if (originalModelDeploy == null) {
-            // NOTE: 修改下面方括号中的话述
             errorMessage = "数据验证失败，当前 [数据] 并不存在，请刷新后重试！";
             return ResponseResult.error(ErrorCodeEnum.DATA_NOT_EXIST, errorMessage);
         }

application-webadmin/src/main/java/com/yice/webadmin/app/controller/ModelInstanceController.java

@@ -12,10 +12,15 @@ import com.yice.common.core.util.MyModelUtil;
 import com.yice.common.core.util.MyPageUtil;
 import com.yice.common.log.annotation.OperationLog;
 import com.yice.common.log.model.constant.SysOperationLogType;
+import com.yice.webadmin.app.config.LlmModelConfig;
 import com.yice.webadmin.app.config.OtherConfig;
+import com.yice.webadmin.app.config.PythonConfig;
+import com.yice.webadmin.app.constant.ModelConstant;
 import com.yice.webadmin.app.dto.ModelInstanceDto;
 import com.yice.webadmin.app.model.ModelInstance;
+import com.yice.webadmin.app.model.ModelVersion;
 import com.yice.webadmin.app.service.ModelInstanceService;
+import com.yice.webadmin.app.service.ModelVersionService;
 import com.yice.webadmin.app.service.ProxyPythonService;
 import com.yice.webadmin.app.vo.ModelInstanceVo;
 import io.swagger.annotations.Api;
@@ -25,6 +30,7 @@ import org.springframework.web.bind.annotation.*;
 
 import java.io.IOException;
 import java.util.List;
+import java.util.stream.Collectors;
 
 /**
  * 模型实例管理操作控制器类。
@@ -44,13 +50,20 @@ public class ModelInstanceController {
     private ProxyPythonService proxyPythonService;
     @Autowired
     private OtherConfig otherConfig;
+    @Autowired
+    private LlmModelConfig llmModelConfig;
+    @Autowired
+    private PythonConfig pythonConfig;
+    @Autowired
+    private ModelVersionService modelVersionService;
+
     /**
      * 新增知识图谱管理数据。
      *
      * @param modelInstanceDto 新增对象。
      * @return 应答结果对象，包含新增对象主键Id。
      */
-    @ApiOperationSupport(ignoreParameters = {"modelInstanceDto.deployId"})
+    @ApiOperationSupport(ignoreParameters = {"modelInstanceDto.instanceId"})
     @OperationLog(type = SysOperationLogType.ADD)
     @PostMapping("/add")
     public ResponseResult<Long> add(@MyRequestBody ModelInstanceDto modelInstanceDto) {
@@ -73,6 +86,113 @@ public class ModelInstanceController {
         return ResponseResult.success(modelInstance.getInstanceId());
     }
 
+    /**
+     * 停止指定LLM模型。
+     *
+     * @return 应答结果对象，包含查询结果集。
+     */
+    @PostMapping("/stop")
+    public ResponseResult<String> stop(@MyRequestBody ModelInstanceDto modelInstanceDto) {
+        ModelInstance modelInstance = modelInstanceService.getById(modelInstanceDto.getInstanceId());
+        ResponseResult<String> responseResult = this.doStop(modelInstance);
+        if (responseResult.isSuccess()) {
+            modelInstance.setDeployStatus(3);
+            modelInstanceService.updateById(modelInstance);
+            return ResponseResult.success(responseResult.getData());
+        }
+        return responseResult;
+    }
+
+    /**
+     * 部署指定LLM模型。
+     *
+     * @return 应答结果对象，包含查询结果集。
+     */
+    @PostMapping("/deploy")
+    public ResponseResult<String> deploy(@MyRequestBody ModelInstanceDto modelInstanceDto,
+                                         @MyRequestBody String type) {
+        ModelInstance modelInstance = modelInstanceService.getById(modelInstanceDto.getInstanceId());
+        modelInstance.setDeployStatus(ModelConstant.DEPLOY_STATUS_DEPLOYED);
+        modelInstanceService.updateById(modelInstance);
+
+        ResponseResult<String> responseResult = this.doReloadOrStart(modelInstance, type);
+        if (responseResult.isSuccess()) {
+            modelInstance.setDeployStatus(ModelConstant.DEPLOY_STATUS_COMPLETE);
+            modelInstanceService.updateById(modelInstance);
+            return ResponseResult.success(responseResult.getData());
+        }
+
+        modelInstance.setDeployStatus(ModelConstant.DEPLOY_STATUS_FAILURE);
+        modelInstanceService.updateById(modelInstance);
+        return responseResult;
+    }
+
+    /**
+     * 调用Python模型方法
+     * @param modelInstance 模型实例
+     * @param type 类型
+     * @return 消息
+     */
+    private ResponseResult<String> doReloadOrStart(ModelInstance modelInstance, String type) {
+        String gps_ids = JSON.parseArray(modelInstance.getResourceInfo()).stream()
+                .map(obj -> ((JSONObject) obj).getString("gpu_id"))
+                .collect(Collectors.joining(","));
+        ModelVersion modelVersion = this.modelVersionService.getById(modelInstance.getVersionId());
+        String requestBody = "{\n" +
+                "  \"new_model_name\": \"" + modelInstance.getVersionName() + "\",\n" +
+                "  \"new_model_path\": \"" + modelVersion.getModelUrl() + "\",\n" +
+                "  \"startup_param\": {\n" +
+                "    \"gpu_ids\": \"" + gps_ids + "\"\n" +
+                "  },\n" +
+                "  \"controller_address\": \"" + pythonConfig.getControllerAddress() + "\"\n" +
+                "}";
+        try {
+            String apiName = type.equals("start") ? llmModelConfig.getStart() : llmModelConfig.getReload();
+            String result = proxyPythonService.predictPost(llmModelConfig.getLlmModelInterface() + apiName, requestBody);
+            JSONObject jo = JSON.parseObject(result);
+            Integer code = jo.getIntValue("code");
+            String msg = jo.getString("msg");
+            String data = jo.getString("data");
+            if (code != null && code == 200) {
+                return ResponseResult.success(data);
+            } else {
+                return ResponseResult.create(ErrorCodeEnum.SERVER_INTERNAL_ERROR, msg, data);
+            }
+        } catch (IOException e) {
+            throw new RuntimeException(e);
+        }
+    }
+
+    /**
+     * 模型停止，调用python接口
+     * @param modelInstance 模型实例
+     * @return 消息
+     */
+    private ResponseResult<String> doStop(ModelInstance modelInstance) {
+        String errorMessage = MyCommonUtil.getModelValidationError(modelInstance, true);
+        if (errorMessage != null) {
+            return ResponseResult.error(ErrorCodeEnum.DATA_VALIDATED_FAILED, errorMessage);
+        }
+        String requestBody = "{\n" +
+                "  \"model_name\": \"" + modelInstance.getVersionName() + "\",\n" +
+                "  \"controller_address\": \"" + pythonConfig.getControllerAddress() + "\"\n" +
+                "}";
+        try {
+            String result = proxyPythonService.predictPost(llmModelConfig.getLlmModelInterface() + llmModelConfig.getStop(), requestBody);
+            JSONObject jo = JSON.parseObject(result);
+            Integer code = jo.getIntValue("code");
+            String msg = jo.getString("msg");
+            String data = jo.getString("data");
+            if (code != null && code == 200) {
+                return ResponseResult.success(data);
+            } else {
+                return ResponseResult.create(ErrorCodeEnum.SERVER_INTERNAL_ERROR, msg, data);
+            }
+        } catch (IOException e) {
+            throw new RuntimeException(e);
+        }
+    }
+
     /**
      * 获取GPU信息。
      *