@dhbw-cloud/llm-client 0.2.17 → 0.2.20

package/dist/index.d.ts

@@ -1,2 +1,2 @@
-export { blockMachine, createKey, deleteCatalogModel, deleteFleetPackage, deleteGpuPool, deleteKey, deleteTreeProject, deleteWorkspaceProject, forgetMachine, getFleet, getFleetInference, getFleetInventoryCsv, getFleetPackage, getFleetProfile, getFleetReadme, getGpuFleet, getGpuHours, getMe, getModelCatalog, getMyWorkspaceUsage, getProjectImageLog, getTreeCatalog, getUsage, getWorkspaceCatalog, getWorkspaceStorage, getWorkspaceUsage, listGpuServers, listKeys, listTreeProjects, listWorkspaceProjects, type Options, putCatalogModel, putGpuNode, putGpuPool, putTierModel, putTreeProject, rebuildProjectImage, setDefaultProject, stopGpuServer, unblockMachine, updateWorkspaceProject } from './sdk.gen';
-export type { AccessRole, BlockMachineData, BlockMachineError, BlockMachineErrors, BlockMachineResponse, BlockMachineResponses, ClientOptions, CreateKeyData, CreateKeyError, CreateKeyErrors, CreateKeyResponse, CreateKeyResponses, DeleteCatalogModelData, DeleteCatalogModelError, DeleteCatalogModelErrors, DeleteCatalogModelResponse, DeleteCatalogModelResponses, DeleteFleetPackageData, DeleteFleetPackageResponse, DeleteFleetPackageResponses, DeleteGpuPoolData, DeleteGpuPoolError, DeleteGpuPoolErrors, DeleteGpuPoolResponse, DeleteGpuPoolResponses, DeleteKeyData, DeleteKeyError, DeleteKeyErrors, DeleteKeyResponse, DeleteKeyResponses, DeleteTreeProjectData, DeleteTreeProjectError, DeleteTreeProjectErrors, DeleteTreeProjectResponse, DeleteTreeProjectResponses, DeleteWorkspaceProjectData, DeleteWorkspaceProjectError, DeleteWorkspaceProjectErrors, DeleteWorkspaceProjectResponse, DeleteWorkspaceProjectResponses, FleetFleetView, FleetPackageInfo, FleetPeerView, FleetProfile, ForgetMachineData, ForgetMachineError, ForgetMachineErrors, ForgetMachineResponse, ForgetMachineResponses, GetFleetData, GetFleetError, GetFleetErrors, GetFleetInferenceData, GetFleetInferenceError, GetFleetInferenceErrors, GetFleetInferenceResponse, GetFleetInferenceResponses, GetFleetInventoryCsvData, GetFleetInventoryCsvResponse, GetFleetInventoryCsvResponses, GetFleetPackageData, GetFleetPackageError, GetFleetPackageErrors, GetFleetPackageResponse, GetFleetPackageResponses, GetFleetProfileData, GetFleetProfileResponse, GetFleetProfileResponses, GetFleetReadmeData, GetFleetReadmeResponse, GetFleetReadmeResponses, GetFleetResponse, GetFleetResponses, GetGpuFleetData, GetGpuFleetError, GetGpuFleetErrors, GetGpuFleetResponse, GetGpuFleetResponses, GetGpuHoursData, GetGpuHoursError, GetGpuHoursErrors, GetGpuHoursResponse, GetGpuHoursResponses, GetMeData, GetMeError, GetMeErrors, GetMeResponse, GetMeResponses, GetModelCatalogData, GetModelCatalogError, GetModelCatalogErrors, GetModelCatalogResponse, GetModelCatalogResponses, GetMyWorkspaceUsageData, GetMyWorkspaceUsageError, GetMyWorkspaceUsageErrors, GetMyWorkspaceUsageResponse, GetMyWorkspaceUsageResponses, GetProjectImageLogData, GetProjectImageLogError, GetProjectImageLogErrors, GetProjectImageLogResponse, GetProjectImageLogResponses, GetTreeCatalogData, GetTreeCatalogError, GetTreeCatalogErrors, GetTreeCatalogResponse, GetTreeCatalogResponses, GetUsageData, GetUsageError, GetUsageErrors, GetUsageResponse, GetUsageResponses, GetWorkspaceCatalogData, GetWorkspaceCatalogError, GetWorkspaceCatalogErrors, GetWorkspaceCatalogResponse, GetWorkspaceCatalogResponses, GetWorkspaceStorageData, GetWorkspaceStorageError, GetWorkspaceStorageErrors, GetWorkspaceStorageResponse, GetWorkspaceStorageResponses, GetWorkspaceUsageData, GetWorkspaceUsageError, GetWorkspaceUsageErrors, GetWorkspaceUsageResponse, GetWorkspaceUsageResponses, GpuCatalogModel, GpuClassInfo, GpuGpuNode, GpuInferencePod, GpuInferencePool, GpuModelCatalog, GpuModelChoice, GpuModelState, GpuNodeAssignment, GpuPool, GpuServer, GpuTier, KeysKeyView, KeysProjectUsage, KeysUsage, ListGpuServersData, ListGpuServersError, ListGpuServersErrors, ListGpuServersResponse, ListGpuServersResponses, ListKeysData, ListKeysError, ListKeysErrors, ListKeysResponse, ListKeysResponses, ListTreeProjectsData, ListTreeProjectsError, ListTreeProjectsErrors, ListTreeProjectsResponse, ListTreeProjectsResponses, ListWorkspaceProjectsData, ListWorkspaceProjectsResponse, ListWorkspaceProjectsResponses, PutCatalogModelData, PutCatalogModelError, PutCatalogModelErrors, PutCatalogModelResponse, PutCatalogModelResponses, PutGpuNodeData, PutGpuNodeError, PutGpuNodeErrors, PutGpuNodeResponse, PutGpuNodeResponses, PutGpuPoolData, PutGpuPoolError, PutGpuPoolErrors, PutGpuPoolResponse, PutGpuPoolResponses, PutTierModelData, PutTierModelError, PutTierModelErrors, PutTierModelResponse, PutTierModelResponses, PutTreeProjectData, PutTreeProjectError, PutTreeProjectErrors, PutTreeProjectResponse, PutTreeProjectResponses, RebuildProjectImageData, RebuildProjectImageError, RebuildProjectImageErrors, RebuildProjectImageResponse, RebuildProjectImageResponses, SetDefaultProjectData, SetDefaultProjectError, SetDefaultProjectErrors, SetDefaultProjectResponse, SetDefaultProjectResponses, StopGpuServerData, StopGpuServerError, StopGpuServerErrors, StopGpuServerResponse, StopGpuServerResponses, UnblockMachineData, UnblockMachineError, UnblockMachineErrors, UnblockMachineResponse, UnblockMachineResponses, UpdateWorkspaceProjectData, UpdateWorkspaceProjectError, UpdateWorkspaceProjectErrors, UpdateWorkspaceProjectResponse, UpdateWorkspaceProjectResponses, UsageGpuHour, UsageGpuHourPerson, UsageGpuReport, UsageReport, UsageTotal, WebserverBuildLogResponse, WebserverCreateKeyRequest, WebserverCreateKeyResponse, WebserverDefaultProjectRequest, WebserverDefaultProjectResponse, WebserverErrorResponse, WebserverGpuFleetResponse, WebserverGpuNodeRequest, WebserverGpuPoolRequest, WebserverMeResponse, WebserverTierModelRequest, WebserverTreeCatalog, WorkspaceBase, WorkspaceBudget, WorkspaceBudgetStatus, WorkspaceConfig, WorkspaceGpuClass, WorkspaceImageBuild, WorkspaceLimits, WorkspaceMember, WorkspaceMine, WorkspaceProfile, WorkspaceProject, WorkspaceStorageDir, WorkspaceStorageReport, WorkspaceTreeBudget, WorkspaceTreeProject, WorkspaceTreeStatus, WorkspaceTreeUse, WorkspaceValkeyMemory } from './types.gen';
+export { blockMachine, createKey, deleteCatalogModel, deleteFleetPackage, deleteGpuPool, deleteKey, deleteTreeProject, deleteWorkspaceProject, forgetMachine, getFleet, getFleetInference, getFleetInventoryCsv, getFleetPackage, getFleetProfile, getFleetReadme, getGpuFleet, getGpuHours, getMe, getModelCatalog, getMyWorkspaceUsage, getProjectImageLog, getTreeCatalog, getUsage, getWorkspaceCatalog, getWorkspaceStorage, getWorkspaceUsage, listGpuServers, listKeys, listTreeProjects, listWorkspaceProjects, type Options, putCatalogModel, putGpuNode, putGpuPool, putTierModel, putTreeProject, rebuildProjectImage, setDefaultProject, setFleetModel, stopGpuServer, unblockMachine, updateWorkspaceProject } from './sdk.gen';
+export type { AccessRole, BlockMachineData, BlockMachineError, BlockMachineErrors, BlockMachineResponse, BlockMachineResponses, ClientOptions, CreateKeyData, CreateKeyError, CreateKeyErrors, CreateKeyResponse, CreateKeyResponses, DeleteCatalogModelData, DeleteCatalogModelError, DeleteCatalogModelErrors, DeleteCatalogModelResponse, DeleteCatalogModelResponses, DeleteFleetPackageData, DeleteFleetPackageResponse, DeleteFleetPackageResponses, DeleteGpuPoolData, DeleteGpuPoolError, DeleteGpuPoolErrors, DeleteGpuPoolResponse, DeleteGpuPoolResponses, DeleteKeyData, DeleteKeyError, DeleteKeyErrors, DeleteKeyResponse, DeleteKeyResponses, DeleteTreeProjectData, DeleteTreeProjectError, DeleteTreeProjectErrors, DeleteTreeProjectResponse, DeleteTreeProjectResponses, DeleteWorkspaceProjectData, DeleteWorkspaceProjectError, DeleteWorkspaceProjectErrors, DeleteWorkspaceProjectResponse, DeleteWorkspaceProjectResponses, FleetFleetView, FleetPackageInfo, FleetPeerView, FleetProfile, ForgetMachineData, ForgetMachineError, ForgetMachineErrors, ForgetMachineResponse, ForgetMachineResponses, GetFleetData, GetFleetError, GetFleetErrors, GetFleetInferenceData, GetFleetInferenceError, GetFleetInferenceErrors, GetFleetInferenceResponse, GetFleetInferenceResponses, GetFleetInventoryCsvData, GetFleetInventoryCsvResponse, GetFleetInventoryCsvResponses, GetFleetPackageData, GetFleetPackageError, GetFleetPackageErrors, GetFleetPackageResponse, GetFleetPackageResponses, GetFleetProfileData, GetFleetProfileResponse, GetFleetProfileResponses, GetFleetReadmeData, GetFleetReadmeResponse, GetFleetReadmeResponses, GetFleetResponse, GetFleetResponses, GetGpuFleetData, GetGpuFleetError, GetGpuFleetErrors, GetGpuFleetResponse, GetGpuFleetResponses, GetGpuHoursData, GetGpuHoursError, GetGpuHoursErrors, GetGpuHoursResponse, GetGpuHoursResponses, GetMeData, GetMeError, GetMeErrors, GetMeResponse, GetMeResponses, GetModelCatalogData, GetModelCatalogError, GetModelCatalogErrors, GetModelCatalogResponse, GetModelCatalogResponses, GetMyWorkspaceUsageData, GetMyWorkspaceUsageError, GetMyWorkspaceUsageErrors, GetMyWorkspaceUsageResponse, GetMyWorkspaceUsageResponses, GetProjectImageLogData, GetProjectImageLogError, GetProjectImageLogErrors, GetProjectImageLogResponse, GetProjectImageLogResponses, GetTreeCatalogData, GetTreeCatalogError, GetTreeCatalogErrors, GetTreeCatalogResponse, GetTreeCatalogResponses, GetUsageData, GetUsageError, GetUsageErrors, GetUsageResponse, GetUsageResponses, GetWorkspaceCatalogData, GetWorkspaceCatalogError, GetWorkspaceCatalogErrors, GetWorkspaceCatalogResponse, GetWorkspaceCatalogResponses, GetWorkspaceStorageData, GetWorkspaceStorageError, GetWorkspaceStorageErrors, GetWorkspaceStorageResponse, GetWorkspaceStorageResponses, GetWorkspaceUsageData, GetWorkspaceUsageError, GetWorkspaceUsageErrors, GetWorkspaceUsageResponse, GetWorkspaceUsageResponses, GpuCatalogModel, GpuClassInfo, GpuGpuNode, GpuInferencePod, GpuInferencePool, GpuModelCatalog, GpuModelChoice, GpuModelState, GpuNodeAssignment, GpuPool, GpuServer, GpuTier, KeysKeyView, KeysProjectUsage, KeysUsage, ListGpuServersData, ListGpuServersError, ListGpuServersErrors, ListGpuServersResponse, ListGpuServersResponses, ListKeysData, ListKeysError, ListKeysErrors, ListKeysResponse, ListKeysResponses, ListTreeProjectsData, ListTreeProjectsError, ListTreeProjectsErrors, ListTreeProjectsResponse, ListTreeProjectsResponses, ListWorkspaceProjectsData, ListWorkspaceProjectsResponse, ListWorkspaceProjectsResponses, PutCatalogModelData, PutCatalogModelError, PutCatalogModelErrors, PutCatalogModelResponse, PutCatalogModelResponses, PutGpuNodeData, PutGpuNodeError, PutGpuNodeErrors, PutGpuNodeResponse, PutGpuNodeResponses, PutGpuPoolData, PutGpuPoolError, PutGpuPoolErrors, PutGpuPoolResponse, PutGpuPoolResponses, PutTierModelData, PutTierModelError, PutTierModelErrors, PutTierModelResponse, PutTierModelResponses, PutTreeProjectData, PutTreeProjectError, PutTreeProjectErrors, PutTreeProjectResponse, PutTreeProjectResponses, RebuildProjectImageData, RebuildProjectImageError, RebuildProjectImageErrors, RebuildProjectImageResponse, RebuildProjectImageResponses, SetDefaultProjectData, SetDefaultProjectError, SetDefaultProjectErrors, SetDefaultProjectResponse, SetDefaultProjectResponses, SetFleetModelData, SetFleetModelError, SetFleetModelErrors, SetFleetModelResponse, SetFleetModelResponses, StopGpuServerData, StopGpuServerError, StopGpuServerErrors, StopGpuServerResponse, StopGpuServerResponses, UnblockMachineData, UnblockMachineError, UnblockMachineErrors, UnblockMachineResponse, UnblockMachineResponses, UpdateWorkspaceProjectData, UpdateWorkspaceProjectError, UpdateWorkspaceProjectErrors, UpdateWorkspaceProjectResponse, UpdateWorkspaceProjectResponses, UsageGpuHour, UsageGpuHourPerson, UsageGpuReport, UsageReport, UsageTotal, WebserverBuildLogResponse, WebserverCreateKeyRequest, WebserverCreateKeyResponse, WebserverDefaultProjectRequest, WebserverDefaultProjectResponse, WebserverErrorResponse, WebserverFleetModelRequest, WebserverFleetModelResponse, WebserverGpuFleetResponse, WebserverGpuNodeRequest, WebserverGpuPoolRequest, WebserverMeResponse, WebserverTierModelRequest, WebserverTreeCatalog, WorkspaceBase, WorkspaceBudget, WorkspaceBudgetStatus, WorkspaceConfig, WorkspaceGpuClass, WorkspaceImageBuild, WorkspaceLimits, WorkspaceMember, WorkspaceMine, WorkspaceProfile, WorkspaceProject, WorkspaceStorageDir, WorkspaceStorageReport, WorkspaceTreeBudget, WorkspaceTreeProject, WorkspaceTreeStatus, WorkspaceTreeUse, WorkspaceValkeyMemory } from './types.gen';

package/dist/index.mjs

@@ -802,6 +802,15 @@ var getFleetInventoryCsv = (options) => (options?.client ?? client).get({
   url: "/v1/fleet/inventory.csv",
   ...options
 });
+var setFleetModel = (options) => (options.client ?? client).put({
+  security: [{ name: "Authorization", type: "apiKey" }],
+  url: "/v1/fleet/model",
+  ...options,
+  headers: {
+    "Content-Type": "application/json",
+    ...options.headers
+  }
+});
 var deleteFleetPackage = (options) => (options?.client ?? client).delete({
   security: [{ name: "Authorization", type: "apiKey" }],
   url: "/v1/fleet/package",
@@ -1058,6 +1067,7 @@ export {
   putTreeProject,
   rebuildProjectImage,
   setDefaultProject,
+  setFleetModel,
   stopGpuServer,
   unblockMachine,
   updateWorkspaceProject

package/dist/sdk.gen.d.ts

@@ -1,5 +1,5 @@
 import type { Client, ClientMeta, Options as Options2, RequestResult, TDataShape } from './client';
-import type { BlockMachineData, BlockMachineErrors, BlockMachineResponses, CreateKeyData, CreateKeyErrors, CreateKeyResponses, DeleteCatalogModelData, DeleteCatalogModelErrors, DeleteCatalogModelResponses, DeleteFleetPackageData, DeleteFleetPackageResponses, DeleteGpuPoolData, DeleteGpuPoolErrors, DeleteGpuPoolResponses, DeleteKeyData, DeleteKeyErrors, DeleteKeyResponses, DeleteTreeProjectData, DeleteTreeProjectErrors, DeleteTreeProjectResponses, DeleteWorkspaceProjectData, DeleteWorkspaceProjectErrors, DeleteWorkspaceProjectResponses, ForgetMachineData, ForgetMachineErrors, ForgetMachineResponses, GetFleetData, GetFleetErrors, GetFleetInferenceData, GetFleetInferenceErrors, GetFleetInferenceResponses, GetFleetInventoryCsvData, GetFleetInventoryCsvResponses, GetFleetPackageData, GetFleetPackageErrors, GetFleetPackageResponses, GetFleetProfileData, GetFleetProfileResponses, GetFleetReadmeData, GetFleetReadmeResponses, GetFleetResponses, GetGpuFleetData, GetGpuFleetErrors, GetGpuFleetResponses, GetGpuHoursData, GetGpuHoursErrors, GetGpuHoursResponses, GetMeData, GetMeErrors, GetMeResponses, GetModelCatalogData, GetModelCatalogErrors, GetModelCatalogResponses, GetMyWorkspaceUsageData, GetMyWorkspaceUsageErrors, GetMyWorkspaceUsageResponses, GetProjectImageLogData, GetProjectImageLogErrors, GetProjectImageLogResponses, GetTreeCatalogData, GetTreeCatalogErrors, GetTreeCatalogResponses, GetUsageData, GetUsageErrors, GetUsageResponses, GetWorkspaceCatalogData, GetWorkspaceCatalogErrors, GetWorkspaceCatalogResponses, GetWorkspaceStorageData, GetWorkspaceStorageErrors, GetWorkspaceStorageResponses, GetWorkspaceUsageData, GetWorkspaceUsageErrors, GetWorkspaceUsageResponses, ListGpuServersData, ListGpuServersErrors, ListGpuServersResponses, ListKeysData, ListKeysErrors, ListKeysResponses, ListTreeProjectsData, ListTreeProjectsErrors, ListTreeProjectsResponses, ListWorkspaceProjectsData, ListWorkspaceProjectsResponses, PutCatalogModelData, PutCatalogModelErrors, PutCatalogModelResponses, PutGpuNodeData, PutGpuNodeErrors, PutGpuNodeResponses, PutGpuPoolData, PutGpuPoolErrors, PutGpuPoolResponses, PutTierModelData, PutTierModelErrors, PutTierModelResponses, PutTreeProjectData, PutTreeProjectErrors, PutTreeProjectResponses, RebuildProjectImageData, RebuildProjectImageErrors, RebuildProjectImageResponses, SetDefaultProjectData, SetDefaultProjectErrors, SetDefaultProjectResponses, StopGpuServerData, StopGpuServerErrors, StopGpuServerResponses, UnblockMachineData, UnblockMachineErrors, UnblockMachineResponses, UpdateWorkspaceProjectData, UpdateWorkspaceProjectErrors, UpdateWorkspaceProjectResponses } from './types.gen';
+import type { BlockMachineData, BlockMachineErrors, BlockMachineResponses, CreateKeyData, CreateKeyErrors, CreateKeyResponses, DeleteCatalogModelData, DeleteCatalogModelErrors, DeleteCatalogModelResponses, DeleteFleetPackageData, DeleteFleetPackageResponses, DeleteGpuPoolData, DeleteGpuPoolErrors, DeleteGpuPoolResponses, DeleteKeyData, DeleteKeyErrors, DeleteKeyResponses, DeleteTreeProjectData, DeleteTreeProjectErrors, DeleteTreeProjectResponses, DeleteWorkspaceProjectData, DeleteWorkspaceProjectErrors, DeleteWorkspaceProjectResponses, ForgetMachineData, ForgetMachineErrors, ForgetMachineResponses, GetFleetData, GetFleetErrors, GetFleetInferenceData, GetFleetInferenceErrors, GetFleetInferenceResponses, GetFleetInventoryCsvData, GetFleetInventoryCsvResponses, GetFleetPackageData, GetFleetPackageErrors, GetFleetPackageResponses, GetFleetProfileData, GetFleetProfileResponses, GetFleetReadmeData, GetFleetReadmeResponses, GetFleetResponses, GetGpuFleetData, GetGpuFleetErrors, GetGpuFleetResponses, GetGpuHoursData, GetGpuHoursErrors, GetGpuHoursResponses, GetMeData, GetMeErrors, GetMeResponses, GetModelCatalogData, GetModelCatalogErrors, GetModelCatalogResponses, GetMyWorkspaceUsageData, GetMyWorkspaceUsageErrors, GetMyWorkspaceUsageResponses, GetProjectImageLogData, GetProjectImageLogErrors, GetProjectImageLogResponses, GetTreeCatalogData, GetTreeCatalogErrors, GetTreeCatalogResponses, GetUsageData, GetUsageErrors, GetUsageResponses, GetWorkspaceCatalogData, GetWorkspaceCatalogErrors, GetWorkspaceCatalogResponses, GetWorkspaceStorageData, GetWorkspaceStorageErrors, GetWorkspaceStorageResponses, GetWorkspaceUsageData, GetWorkspaceUsageErrors, GetWorkspaceUsageResponses, ListGpuServersData, ListGpuServersErrors, ListGpuServersResponses, ListKeysData, ListKeysErrors, ListKeysResponses, ListTreeProjectsData, ListTreeProjectsErrors, ListTreeProjectsResponses, ListWorkspaceProjectsData, ListWorkspaceProjectsResponses, PutCatalogModelData, PutCatalogModelErrors, PutCatalogModelResponses, PutGpuNodeData, PutGpuNodeErrors, PutGpuNodeResponses, PutGpuPoolData, PutGpuPoolErrors, PutGpuPoolResponses, PutTierModelData, PutTierModelErrors, PutTierModelResponses, PutTreeProjectData, PutTreeProjectErrors, PutTreeProjectResponses, RebuildProjectImageData, RebuildProjectImageErrors, RebuildProjectImageResponses, SetDefaultProjectData, SetDefaultProjectErrors, SetDefaultProjectResponses, SetFleetModelData, SetFleetModelErrors, SetFleetModelResponses, StopGpuServerData, StopGpuServerErrors, StopGpuServerResponses, UnblockMachineData, UnblockMachineErrors, UnblockMachineResponses, UpdateWorkspaceProjectData, UpdateWorkspaceProjectErrors, UpdateWorkspaceProjectResponses } from './types.gen';
 export type Options<TData extends TDataShape = TDataShape, ThrowOnError extends boolean = boolean, TResponse = unknown> = Options2<TData, ThrowOnError, TResponse> & {
     /**
      * You can provide a client instance returned by `createClient()` instead of
@@ -29,6 +29,12 @@ export declare const getFleetInference: <ThrowOnError extends boolean = false>(o
  * Fleet inventory as CSV
  */
 export declare const getFleetInventoryCsv: <ThrowOnError extends boolean = false>(options?: Options<GetFleetInventoryCsvData, ThrowOnError>) => RequestResult<GetFleetInventoryCsvResponses, unknown, ThrowOnError>;
+/**
+ * Set the model of machines
+ *
+ * The model wins over the machine's profile (by memory). A machine pulls it at its next enrolment, within reenrollSeconds; context and memory limit stay the profile's. An empty model returns the machines to their profile's.
+ */
+export declare const setFleetModel: <ThrowOnError extends boolean = false>(options: Options<SetFleetModelData, ThrowOnError>) => RequestResult<SetFleetModelResponses, SetFleetModelErrors, ThrowOnError>;
 /**
  * Withdraw the uploaded JAMF package
  *

package/dist/types.gen.d.ts

@@ -52,6 +52,14 @@ export type FleetPeerView = {
     last_seen?: string;
     location?: string;
     model?: string;
+    /**
+     * why the machine could not pull or load its model
+     */
+    model_error?: string;
+    /**
+     * set by an admin; empty = the profile's
+     */
+    model_override?: string;
     model_state?: 'fremdverwaltet' | 'erfüllt' | 'abweichend' | 'unbekannt';
     models?: Array<string>;
     name?: string;
@@ -84,6 +92,10 @@ export type GpuCatalogModel = {
      * vLLM arguments after --model and --port
      */
     args?: Array<string>;
+    /**
+     * Category is what it is for, as stated: chat, code, fast or general. The measured one (ModelState.Category) wins once the model has run.
+     */
+    category?: 'chat' | 'code' | 'fast' | 'general';
     /**
      * Resources of the replica's container; empty = 1 CPU, 3–5 GiB, 1 GPU.
      */
@@ -108,6 +120,10 @@ export type GpuClassInfo = {
      */
     gpus?: number;
     id?: string;
+    /**
+     * its pool lends idle GPUs to the other pools of the tier
+     */
+    lends?: boolean;
     name?: string;
     /**
      * assigned nodes, ready or not
@@ -193,15 +209,24 @@ export type GpuModelCatalog = {
     tiers?: Array<GpuTier>;
 };
 export type GpuModelChoice = {
+    /**
+     * measured, else stated
+     */
+    category?: string;
     key?: string;
     min_gb?: number;
 };
 export type GpuModelState = {
+    /**
+     * Category measured by the alias classification (code before chat before fast) once the model ran; empty before. Reason says why.
+     */
+    category?: 'chat' | 'code' | 'fast';
     /**
      * the last failure, e.g. CrashLoopBackOff: Error
      */
     problem?: string;
     ready?: number;
+    reason?: string;
     replicas?: number;
     restarts?: number;
     starting?: number;
@@ -226,7 +251,7 @@ export type GpuPool = {
     created_at?: string;
     id?: string;
     /**
-     * idle GPUs of this pool may be used by the other pools of the tier (not yet in effect)
+     * idle GPUs of this pool may be used by the other pools of the tier
      */
     lend?: boolean;
     name?: string;
@@ -391,6 +416,16 @@ export type WebserverErrorResponse = {
      */
     error?: string;
 };
+export type WebserverFleetModelRequest = {
+    /**
+     * Ollama model name; empty = back to the profile's model.
+     */
+    model?: string;
+    serials: Array<string>;
+};
+export type WebserverFleetModelResponse = {
+    changed?: number;
+};
 export type WebserverGpuFleetResponse = {
     classes?: Array<GpuClassInfo>;
     /**
@@ -499,6 +534,10 @@ export type WorkspaceGpuClass = {
      */
     gpus?: number;
     id?: string;
+    /**
+     * Lends: its pool lends idle GPUs to the other pools of the tier.
+     */
+    lends?: boolean;
     name?: string;
     nodes?: number;
     pool?: string;
@@ -812,6 +851,33 @@ export type GetFleetInventoryCsvResponses = {
     200: string;
 };
 export type GetFleetInventoryCsvResponse = GetFleetInventoryCsvResponses[keyof GetFleetInventoryCsvResponses];
+export type SetFleetModelData = {
+    /**
+     * machines and model
+     */
+    body: WebserverFleetModelRequest;
+    path?: never;
+    query?: never;
+    url: '/v1/fleet/model';
+};
+export type SetFleetModelErrors = {
+    /**
+     * Bad Request
+     */
+    400: WebserverErrorResponse;
+    /**
+     * Not Found
+     */
+    404: WebserverErrorResponse;
+};
+export type SetFleetModelError = SetFleetModelErrors[keyof SetFleetModelErrors];
+export type SetFleetModelResponses = {
+    /**
+     * OK
+     */
+    200: WebserverFleetModelResponse;
+};
+export type SetFleetModelResponse = SetFleetModelResponses[keyof SetFleetModelResponses];
 export type DeleteFleetPackageData = {
     body?: never;
     path?: never;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dhbw-cloud/llm-client",
-  "version": "0.2.17",
+  "version": "0.2.20",
   "description": "Generated TypeScript client for the DHBW Cloud llm-management-api.",
   "license": "Apache-2.0",
   "type": "module",