@daocloud-proto/hydra 0.16.0 → 0.17.0-rc1-dev-1

package/admin-api/auto-route/v1alpha1/auto_route.pb.ts

@@ -0,0 +1,338 @@
+/* eslint-disable */
+// @ts-nocheck
+/*
+* This file is a generated Typescript file for GRPC Gateway, DO NOT MODIFY
+*/
+
+import * as HydraCommonCommon from "../../../common/common.pb"
+import * as fm from "../../../fetch.pb"
+import * as GoogleProtobufEmpty from "../../../google/protobuf/empty.pb"
+import * as GoogleProtobufTimestamp from "../../../google/protobuf/timestamp.pb"
+
+type Absent<T, K extends keyof T> = { [k in Exclude<keyof T, K>]?: undefined };
+type OneOf<T> =
+  | { [k in keyof T]?: undefined }
+  | (
+    keyof T extends infer K ?
+      (K extends string & keyof T ? { [k in K]: T[K] } & Absent<T, K>
+        : never)
+    : never);
+
+export enum AutoRoutePhase {
+  AUTO_ROUTE_PHASE_UNSPECIFIED = "AUTO_ROUTE_PHASE_UNSPECIFIED",
+  AUTO_ROUTE_PHASE_DISABLED = "AUTO_ROUTE_PHASE_DISABLED",
+  AUTO_ROUTE_PHASE_NOT_READY = "AUTO_ROUTE_PHASE_NOT_READY",
+  AUTO_ROUTE_PHASE_READY = "AUTO_ROUTE_PHASE_READY",
+  AUTO_ROUTE_PHASE_ERROR = "AUTO_ROUTE_PHASE_ERROR",
+}
+
+export enum Strategy {
+  STRATEGY_UNSPECIFIED = "STRATEGY_UNSPECIFIED",
+  STRATEGY_VSR = "STRATEGY_VSR",
+  STRATEGY_WASM = "STRATEGY_WASM",
+}
+
+export enum RouterModelSource {
+  ROUTER_MODEL_SOURCE_UNSPECIFIED = "ROUTER_MODEL_SOURCE_UNSPECIFIED",
+  ROUTER_MODEL_SOURCE_DATASET = "ROUTER_MODEL_SOURCE_DATASET",
+  ROUTER_MODEL_SOURCE_USER_PVC = "ROUTER_MODEL_SOURCE_USER_PVC",
+}
+
+export enum ClassifyCategory {
+  CLASSIFY_CATEGORY_UNSPECIFIED = "CLASSIFY_CATEGORY_UNSPECIFIED",
+  CLASSIFY_CATEGORY_MATH = "CLASSIFY_CATEGORY_MATH",
+  CLASSIFY_CATEGORY_PHYSICS = "CLASSIFY_CATEGORY_PHYSICS",
+  CLASSIFY_CATEGORY_CHEMISTRY = "CLASSIFY_CATEGORY_CHEMISTRY",
+  CLASSIFY_CATEGORY_BIOLOGY = "CLASSIFY_CATEGORY_BIOLOGY",
+  CLASSIFY_CATEGORY_COMPUTER_SCIENCE = "CLASSIFY_CATEGORY_COMPUTER_SCIENCE",
+  CLASSIFY_CATEGORY_HISTORY = "CLASSIFY_CATEGORY_HISTORY",
+  CLASSIFY_CATEGORY_ECONOMICS = "CLASSIFY_CATEGORY_ECONOMICS",
+  CLASSIFY_CATEGORY_BUSINESS = "CLASSIFY_CATEGORY_BUSINESS",
+  CLASSIFY_CATEGORY_LAW = "CLASSIFY_CATEGORY_LAW",
+  CLASSIFY_CATEGORY_HEALTH = "CLASSIFY_CATEGORY_HEALTH",
+  CLASSIFY_CATEGORY_PSYCHOLOGY = "CLASSIFY_CATEGORY_PSYCHOLOGY",
+  CLASSIFY_CATEGORY_PHILOSOPHY = "CLASSIFY_CATEGORY_PHILOSOPHY",
+  CLASSIFY_CATEGORY_ENGINEERING = "CLASSIFY_CATEGORY_ENGINEERING",
+  CLASSIFY_CATEGORY_OTHER = "CLASSIFY_CATEGORY_OTHER",
+  CLASSIFY_CATEGORY_UNKNOWN = "CLASSIFY_CATEGORY_UNKNOWN",
+}
+
+export enum SignalKind {
+  SIGNAL_KIND_UNSPECIFIED = "SIGNAL_KIND_UNSPECIFIED",
+  SIGNAL_KIND_CONTEXT = "SIGNAL_KIND_CONTEXT",
+  SIGNAL_KIND_EMBEDDING = "SIGNAL_KIND_EMBEDDING",
+  SIGNAL_KIND_COMPLEXITY = "SIGNAL_KIND_COMPLEXITY",
+  SIGNAL_KIND_DOMAIN = "SIGNAL_KIND_DOMAIN",
+}
+
+export enum ComplexityLevel {
+  COMPLEXITY_LEVEL_UNSPECIFIED = "COMPLEXITY_LEVEL_UNSPECIFIED",
+  COMPLEXITY_LEVEL_EASY = "COMPLEXITY_LEVEL_EASY",
+  COMPLEXITY_LEVEL_MEDIUM = "COMPLEXITY_LEVEL_MEDIUM",
+  COMPLEXITY_LEVEL_HARD = "COMPLEXITY_LEVEL_HARD",
+}
+
+export enum SignalCombinationOperator {
+  SIGNAL_COMBINATION_OPERATOR_UNSPECIFIED = "SIGNAL_COMBINATION_OPERATOR_UNSPECIFIED",
+  SIGNAL_COMBINATION_OPERATOR_AND = "SIGNAL_COMBINATION_OPERATOR_AND",
+  SIGNAL_COMBINATION_OPERATOR_OR = "SIGNAL_COMBINATION_OPERATOR_OR",
+  SIGNAL_COMBINATION_OPERATOR_NOT = "SIGNAL_COMBINATION_OPERATOR_NOT",
+}
+
+export enum AggregationMethod {
+  AGGREGATION_METHOD_UNSPECIFIED = "AGGREGATION_METHOD_UNSPECIFIED",
+  AGGREGATION_METHOD_MEAN = "AGGREGATION_METHOD_MEAN",
+  AGGREGATION_METHOD_MAX = "AGGREGATION_METHOD_MAX",
+  AGGREGATION_METHOD_ANY = "AGGREGATION_METHOD_ANY",
+}
+
+export type AutoRoute = {
+  cluster?: string
+  spec?: AutoRouteSpec
+  createdAt?: GoogleProtobufTimestamp.Timestamp
+  updatedAt?: GoogleProtobufTimestamp.Timestamp
+  status?: AutoRouteStatus
+}
+
+export type AutoRouteStatus = {
+  phase?: AutoRoutePhase
+  message?: string
+  vsrPhase?: string
+  readyReplicas?: number
+  entrypoints?: EntrypointStatus
+  modelAssets?: ModelAssetStatus[]
+  strategy?: Strategy
+}
+
+export type EntrypointStatus = {
+  envoyfilterReady?: boolean
+  gatewayReady?: boolean
+}
+
+export type ModelAssetStatus = {
+  signalClass?: string
+  source?: RouterModelSource
+  name?: string
+  phase?: string
+  modelPath?: string
+}
+
+
+type BaseAutoRouteSpec = {
+  strategy?: Strategy
+  vsr?: VSRConfig
+  wasm?: WASMConfig
+}
+
+export type AutoRouteSpec = BaseAutoRouteSpec
+  & OneOf<{ enabled: boolean }>
+
+export type VSRConfig = {
+  context?: ContextSignals
+  embeddings?: EmbeddingSignals
+  complexity?: ComplexitySignals
+  classify?: ClassifySignals
+  decisions?: RouteDecision[]
+  fallback?: RouteTarget
+}
+
+export type WASMConfig = {
+  rules?: WASMRule[]
+  fallback?: RouteTarget
+}
+
+export type WASMRule = {
+  name?: string
+  enabled?: boolean
+  description?: string
+  expression?: string
+  target?: RouteTarget
+}
+
+export type ContextSignals = {
+  enabled?: boolean
+  rules?: ContextSignal[]
+}
+
+export type EmbeddingSignals = {
+  enabled?: boolean
+  model?: RouterModelDownload
+  rules?: EmbeddingSignal[]
+}
+
+export type ComplexitySignals = {
+  enabled?: boolean
+  model?: RouterModelDownload
+  rules?: ComplexitySignal[]
+}
+
+export type ClassifySignals = {
+  enabled?: boolean
+  model?: RouterModelDownload
+  rules?: ClassifySignal[]
+}
+
+export type RouterModelDownload = {
+  source?: RouterModelSource
+  modelName?: string
+  pvcName?: string
+}
+
+export type ContextSignal = {
+  name?: string
+  minTokens?: number
+  maxTokens?: number
+  description?: string
+}
+
+export type EmbeddingSignal = {
+  name?: string
+  threshold?: number
+  candidates?: string[]
+  description?: string
+  aggregationMethod?: AggregationMethod
+}
+
+export type ComplexitySignal = {
+  name?: string
+  hardCandidates?: string[]
+  easyCandidates?: string[]
+  threshold?: number
+  description?: string
+}
+
+export type ClassifySignal = {
+  name?: string
+  categories?: ClassifyCategory[]
+  description?: string
+}
+
+export type RouteDecision = {
+  name?: string
+  priority?: number
+  operator?: SignalCombinationOperator
+  signalRefs?: RouteSignalRef[]
+  target?: RouteTarget
+  description?: string
+}
+
+export type RouteTarget = {
+  maasModelId?: string
+  customModelName?: string
+}
+
+export type RouteSignalRef = {
+  name?: string
+  kind?: SignalKind
+  complexityLevel?: ComplexityLevel
+}
+
+export type CreateAutoRouteRequest = {
+  cluster?: string
+  spec?: AutoRouteSpec
+}
+
+export type GetAutoRouteRequest = {
+  cluster?: string
+  strategy?: Strategy
+}
+
+export type ListAutoRoutesRequest = {
+  page?: HydraCommonCommon.Pagination
+  strategy?: Strategy
+}
+
+export type ListAutoRoutesResponse = {
+  items?: AutoRoute[]
+  page?: HydraCommonCommon.Pagination
+}
+
+export type UpdateAutoRouteRequest = {
+  cluster?: string
+  spec?: AutoRouteSpec
+}
+
+export type UpdateAutoRouteStatusRequest = {
+  cluster?: string
+  enabled?: boolean
+  strategy?: Strategy
+}
+
+export type DeleteAutoRouteRequest = {
+  cluster?: string
+  strategy?: Strategy
+}
+
+export type RouterModelConfig = {
+  category?: string
+  modelType?: string
+  repoId?: string
+  modelPath?: string
+}
+
+export type ListRouterModelsRequest = {
+  page?: HydraCommonCommon.Pagination
+}
+
+export type ListRouterModelsResponse = {
+  items?: RouterModelConfig[]
+  page?: HydraCommonCommon.Pagination
+}
+
+export type ModelServing = {
+  name?: string
+  accessModelName?: string
+}
+
+export type ListModelServingsRequest = {
+  cluster?: string
+  page?: HydraCommonCommon.Pagination
+}
+
+export type ListModelServingsResponse = {
+  items?: ModelServing[]
+  page?: HydraCommonCommon.Pagination
+}
+
+export type CheckUserPVCModelRequest = {
+  cluster?: string
+  pvcName?: string
+  modelName?: string
+}
+
+export type CheckUserPVCModelResponse = {
+  found?: boolean
+  configFound?: boolean
+  expectedPath?: string
+  message?: string
+}
+
+export class AutoRouteManagement {
+  static CreateAutoRoute(req: CreateAutoRouteRequest, initReq?: fm.InitReq): Promise<AutoRoute> {
+    return fm.fetchReq<CreateAutoRouteRequest, AutoRoute>(`/apis/admin.hydra.io/v1alpha1/auto-routes`, {...initReq, method: "POST", body: JSON.stringify(req, fm.replacer)})
+  }
+  static GetAutoRoute(req: GetAutoRouteRequest, initReq?: fm.InitReq): Promise<AutoRoute> {
+    return fm.fetchReq<GetAutoRouteRequest, AutoRoute>(`/apis/admin.hydra.io/v1alpha1/auto-routes/${req["cluster"]}?${fm.renderURLSearchParams(req, ["cluster"])}`, {...initReq, method: "GET"})
+  }
+  static ListAutoRoutes(req: ListAutoRoutesRequest, initReq?: fm.InitReq): Promise<ListAutoRoutesResponse> {
+    return fm.fetchReq<ListAutoRoutesRequest, ListAutoRoutesResponse>(`/apis/admin.hydra.io/v1alpha1/auto-routes?${fm.renderURLSearchParams(req, [])}`, {...initReq, method: "GET"})
+  }
+  static UpdateAutoRoute(req: UpdateAutoRouteRequest, initReq?: fm.InitReq): Promise<AutoRoute> {
+    return fm.fetchReq<UpdateAutoRouteRequest, AutoRoute>(`/apis/admin.hydra.io/v1alpha1/auto-routes/${req["cluster"]}`, {...initReq, method: "PUT", body: JSON.stringify(req, fm.replacer)})
+  }
+  static UpdateAutoRouteStatus(req: UpdateAutoRouteStatusRequest, initReq?: fm.InitReq): Promise<AutoRoute> {
+    return fm.fetchReq<UpdateAutoRouteStatusRequest, AutoRoute>(`/apis/admin.hydra.io/v1alpha1/auto-routes/${req["cluster"]}/status`, {...initReq, method: "PUT", body: JSON.stringify(req, fm.replacer)})
+  }
+  static DeleteAutoRoute(req: DeleteAutoRouteRequest, initReq?: fm.InitReq): Promise<GoogleProtobufEmpty.Empty> {
+    return fm.fetchReq<DeleteAutoRouteRequest, GoogleProtobufEmpty.Empty>(`/apis/admin.hydra.io/v1alpha1/auto-routes/${req["cluster"]}`, {...initReq, method: "DELETE"})
+  }
+  static ListRouterModels(req: ListRouterModelsRequest, initReq?: fm.InitReq): Promise<ListRouterModelsResponse> {
+    return fm.fetchReq<ListRouterModelsRequest, ListRouterModelsResponse>(`/apis/admin.hydra.io/v1alpha1/auto-route-models?${fm.renderURLSearchParams(req, [])}`, {...initReq, method: "GET"})
+  }
+  static ListModelServings(req: ListModelServingsRequest, initReq?: fm.InitReq): Promise<ListModelServingsResponse> {
+    return fm.fetchReq<ListModelServingsRequest, ListModelServingsResponse>(`/apis/admin.hydra.io/v1alpha1/auto-routes/${req["cluster"]}/model-servings?${fm.renderURLSearchParams(req, ["cluster"])}`, {...initReq, method: "GET"})
+  }
+  static CheckUserPVCModel(req: CheckUserPVCModelRequest, initReq?: fm.InitReq): Promise<CheckUserPVCModelResponse> {
+    return fm.fetchReq<CheckUserPVCModelRequest, CheckUserPVCModelResponse>(`/apis/admin.hydra.io/v1alpha1/auto-routes/${req["cluster"]}/check-pvc-model`, {...initReq, method: "POST", body: JSON.stringify(req, fm.replacer)})
+  }
+}

package/admin-api/core/v1alpha1/core.pb.ts

@@ -103,6 +103,26 @@ export type ListModelGPUInfoResponse = {
   items?: ModelGPUInfo[]
 }
 
+export type AdminListPersistentVolumeClaimsRequest = {
+  cluster?: string
+  page?: HydraCommonCommon.Pagination
+}
+
+export type PersistentVolumeClaim = {
+  name?: string
+  namespace?: string
+  storageClassName?: string
+  phase?: string
+  capacity?: string
+  accessModes?: string[]
+  createdAt?: string
+}
+
+export type AdminListPersistentVolumeClaimsResponse = {
+  items?: PersistentVolumeClaim[]
+  page?: HydraCommonCommon.Pagination
+}
+
 export class CoreService {
   static AdminListCluster(req: ListClusterRequest, initReq?: fm.InitReq): Promise<ListClusterResponse> {
     return fm.fetchReq<ListClusterRequest, ListClusterResponse>(`/apis/admin.hydra.io/v1alpha1/clusters?${fm.renderURLSearchParams(req, [])}`, {...initReq, method: "GET"})
@@ -125,4 +145,7 @@ export class CoreService {
   static ListModelGPUInfo(req: ListModelGPUInfoRequest, initReq?: fm.InitReq): Promise<ListModelGPUInfoResponse> {
     return fm.fetchReq<ListModelGPUInfoRequest, ListModelGPUInfoResponse>(`/apis/admin.hydra.io/v1alpha1/model-gpu-info?${fm.renderURLSearchParams(req, [])}`, {...initReq, method: "GET"})
   }
+  static AdminListPersistentVolumeClaims(req: AdminListPersistentVolumeClaimsRequest, initReq?: fm.InitReq): Promise<AdminListPersistentVolumeClaimsResponse> {
+    return fm.fetchReq<AdminListPersistentVolumeClaimsRequest, AdminListPersistentVolumeClaimsResponse>(`/apis/admin.hydra.io/v1alpha1/clusters/${req["cluster"]}/persistentvolumeclaims?${fm.renderURLSearchParams(req, ["cluster"])}`, {...initReq, method: "GET"})
+  }
 }

package/admin-api/cost-mock-config/v1alpha1/cost_mock_config.pb.ts

@@ -5,6 +5,15 @@
 */
 
 import * as fm from "../../../fetch.pb"
+import * as GoogleProtobufTimestamp from "../../../google/protobuf/timestamp.pb"
+
+export enum GPUCostDepreciationStatus {
+  DEPRECIATION_STATUS_UNSPECIFIED = "DEPRECIATION_STATUS_UNSPECIFIED",
+  PENDING = "PENDING",
+  ACTIVE = "ACTIVE",
+  EXPIRED = "EXPIRED",
+}
+
 export type CostMockConfig = {
   gpus?: GPUCost[]
   upstreamTokenPrices?: UpstreamTokenPrice[]
@@ -19,6 +28,12 @@ export type GPUCost = {
   memory?: string
   price?: number
   count?: number
+  purchasePrice?: number
+  depreciationStartTime?: GoogleProtobufTimestamp.Timestamp
+  usefulLifeMonths?: number
+  residualValue?: number
+  depreciationStatus?: GPUCostDepreciationStatus
+  hourlyDepreciation?: number
 }
 
 export type UpstreamTokenPrice = {

package/admin-api/guardrails/v1alpha1/guardrails.pb.ts

@@ -150,6 +150,39 @@ export type QueryAIAuditLogResponse = {
   page?: HydraCommonCommon.Pagination
 }
 
+export type QueryAIAuditLogStatisticsRequest = {
+  startTime?: GoogleProtobufTimestamp.Timestamp
+  endTime?: GoogleProtobufTimestamp.Timestamp
+  pluginId?: GuardrailsPlugin
+  level?: GuardrailsLevel
+  routeName?: string
+}
+
+export type AIAuditLogPolicyCount = {
+  pluginId?: GuardrailsPlugin
+  count?: string
+}
+
+export type AIAuditLogRouteNameCount = {
+  routeName?: string
+  count?: string
+}
+
+export type AIAuditLogRouteStatistics = {
+  routeName?: string
+  riskRequestCount?: string
+  logCount?: string
+  blockedCount?: string
+  policyCounts?: AIAuditLogPolicyCount[]
+  averageProcessingDurationMs?: number
+}
+
+export type QueryAIAuditLogStatisticsResponse = {
+  statisticsStartTime?: GoogleProtobufTimestamp.Timestamp
+  statisticsEndTime?: GoogleProtobufTimestamp.Timestamp
+  routeStatistics?: AIAuditLogRouteStatistics[]
+}
+
 export type ExportAIAuditLogRequest = {
   startTime?: GoogleProtobufTimestamp.Timestamp
   endTime?: GoogleProtobufTimestamp.Timestamp
@@ -174,6 +207,9 @@ export class AIGuardrailsService {
   static QueryAIAuditLog(req: QueryAIAuditLogRequest, initReq?: fm.InitReq): Promise<QueryAIAuditLogResponse> {
     return fm.fetchReq<QueryAIAuditLogRequest, QueryAIAuditLogResponse>(`/apis/admin.hydra.io/v1alpha1/ai-guardrails-logs?${fm.renderURLSearchParams(req, [])}`, {...initReq, method: "GET"})
   }
+  static QueryAIAuditLogStatistics(req: QueryAIAuditLogStatisticsRequest, initReq?: fm.InitReq): Promise<QueryAIAuditLogStatisticsResponse> {
+    return fm.fetchReq<QueryAIAuditLogStatisticsRequest, QueryAIAuditLogStatisticsResponse>(`/apis/admin.hydra.io/v1alpha1/ai-guardrails-logs/statistics?${fm.renderURLSearchParams(req, [])}`, {...initReq, method: "GET"})
+  }
   static ExportAIAuditLog(req: ExportAIAuditLogRequest, initReq?: fm.InitReq): Promise<GoogleApiHttpbody.HttpBody> {
     return fm.fetchReq<ExportAIAuditLogRequest, GoogleApiHttpbody.HttpBody>(`/apis/admin.hydra.io/v1alpha1/ai-guardrails-logs/export?${fm.renderURLSearchParams(req, [])}`, {...initReq, method: "GET"})
   }

package/admin-api/maas/v1alpha1/maas.pb.ts

@@ -9,6 +9,12 @@ import * as fm from "../../../fetch.pb"
 import * as GoogleProtobufTimestamp from "../../../google/protobuf/timestamp.pb"
 import * as HydraAdmin_apiModelV1alpha1Model from "../../model/v1alpha1/model.pb"
 
+export enum MAASWorkspaceVisibilityScope {
+  MAAS_WORKSPACE_VISIBILITY_SCOPE_UNSPECIFIED = "MAAS_WORKSPACE_VISIBILITY_SCOPE_UNSPECIFIED",
+  MAAS_WORKSPACE_VISIBILITY_SCOPE_ALL = "MAAS_WORKSPACE_VISIBILITY_SCOPE_ALL",
+  MAAS_WORKSPACE_VISIBILITY_SCOPE_SPECIFIED = "MAAS_WORKSPACE_VISIBILITY_SCOPE_SPECIFIED",
+}
+
 export enum LoadBalanceStrategy {
   LOAD_BALANCE_STRATEGY_UNSPECIFIED = "LOAD_BALANCE_STRATEGY_UNSPECIFIED",
   LOAD_BALANCE_STRATEGY_ROUND_ROBIN = "LOAD_BALANCE_STRATEGY_ROUND_ROBIN",
@@ -22,6 +28,12 @@ export enum RateLimitBaseOn {
   RATE_LIMIT_BASE_ON_WS = "RATE_LIMIT_BASE_ON_WS",
 }
 
+export enum MAASBackendStatus {
+  MAAS_BACKEND_STATUS_UNSPECIFIED = "MAAS_BACKEND_STATUS_UNSPECIFIED",
+  MAAS_BACKEND_STATUS_HEALTHY = "MAAS_BACKEND_STATUS_HEALTHY",
+  MAAS_BACKEND_STATUS_FAILED = "MAAS_BACKEND_STATUS_FAILED",
+}
+
 export enum MAASModelGatewayStatus {
   GATEWAY_STATUS_UNSPECIFIED = "GATEWAY_STATUS_UNSPECIFIED",
   GATEWAY_STATUS_HEALTHY = "GATEWAY_STATUS_HEALTHY",
@@ -34,6 +46,11 @@ export enum UpstreamModelStatus {
   STATUS_FAILED = "STATUS_FAILED",
 }
 
+export type MAASWorkspaceVisibility = {
+  scope?: MAASWorkspaceVisibilityScope
+  workspaces?: number[]
+}
+
 export type CreateMAASModelRequest = {
   modelId?: string
   upstreamModels?: UpstreamModel[]
@@ -41,6 +58,8 @@ export type CreateMAASModelRequest = {
   ratelimitConfig?: RatelimitConfig
   loadBalanceConfig?: LoadBalanceConfig
   tokenWeight?: string
+  workspaceVisibility?: MAASWorkspaceVisibility
+  capabilityRoutes?: MAASCapabilityRoute[]
 }
 
 export type MAASModel = {
@@ -53,6 +72,8 @@ export type MAASModel = {
   gatewayStatus?: MAASModelGatewayStatus
   enabled?: boolean
   tokenWeight?: string
+  workspaceVisibility?: MAASWorkspaceVisibility
+  capabilityRoutes?: MAASCapabilityRoute[]
 }
 
 export type LoadBalanceConfig = {
@@ -82,6 +103,19 @@ export type RateLimitRule = {
   apiKey?: string
 }
 
+export type MAASCapabilityRoute = {
+  capabilities?: string[]
+  upstreams?: MAASCapabilityUpstream[]
+}
+
+export type MAASCapabilityUpstream = {
+  modelId?: string
+  baseUrl?: string
+  capabilityPaths?: {[key: string]: string}
+  apiKey?: string
+  status?: MAASBackendStatus
+}
+
 export type UpstreamModel = {
   modelId?: string
   apiKey?: string
@@ -104,6 +138,8 @@ export type UpdateMAASModelRequest = {
   loadBalanceConfig?: LoadBalanceConfig
   upstreamModels?: UpstreamModel[]
   tokenWeight?: string
+  workspaceVisibility?: MAASWorkspaceVisibility
+  capabilityRoutes?: MAASCapabilityRoute[]
 }
 
 export type UpdateMAASModelStatusRequest = {

package/admin-api/model/v1alpha1/model.pb.ts

@@ -40,6 +40,9 @@ export type CreateModelRequest = {
   servingSpec?: HydraManagement_apiModelV1alpha1Model.ModelServingSpec
   hidden?: boolean
   providerName?: HydraCommonCommon.I18nName
+  parameterSize?: number
+  contextLength?: HydraManagement_apiModelV1alpha1Model.ContextLengthRange
+  operationTags?: HydraManagement_apiModelV1alpha1Model.ModelOperationTag[]
 }
 
 export type Model = {
@@ -59,6 +62,10 @@ export type Model = {
   publicModelConfig?: HydraManagement_apiModelV1alpha1Model.PublicModelConfig
   servingSpec?: HydraManagement_apiModelV1alpha1Model.ModelServingSpec
   hidden?: boolean
+  parameterSize?: number
+  contextLength?: HydraManagement_apiModelV1alpha1Model.ContextLengthRange
+  operationTags?: HydraManagement_apiModelV1alpha1Model.ModelOperationTag[]
+  publishTimestamp?: GoogleProtobufTimestamp.Timestamp
 }
 
 export type UpdateModelRequest = {
@@ -75,6 +82,9 @@ export type UpdateModelRequest = {
   publicModelConfig?: HydraManagement_apiModelV1alpha1Model.PublicModelConfig
   servingSpec?: HydraManagement_apiModelV1alpha1Model.ModelServingSpec
   hidden?: boolean
+  parameterSize?: number
+  contextLength?: HydraManagement_apiModelV1alpha1Model.ContextLengthRange
+  operationTags?: HydraManagement_apiModelV1alpha1Model.ModelOperationTag[]
 }
 
 export type ListModelRequest = {
@@ -157,6 +167,10 @@ export type ListModelServingsByModelResponse = {
   items?: WorkspaceModelServingGroup[]
 }
 
+export type SortModelsRequest = {
+  modelIds?: string[]
+}
+
 export class AdminModelManagement {
   static CreateModel(req: CreateModelRequest, initReq?: fm.InitReq): Promise<Model> {
     return fm.fetchReq<CreateModelRequest, Model>(`/apis/admin.hydra.io/v1alpha1/models`, {...initReq, method: "POST", body: JSON.stringify(req, fm.replacer)})
@@ -194,6 +208,9 @@ export class AdminModelManagement {
   static ListModelServingsByModel(req: ListModelServingsByModelRequest, initReq?: fm.InitReq): Promise<ListModelServingsByModelResponse> {
     return fm.fetchReq<ListModelServingsByModelRequest, ListModelServingsByModelResponse>(`/apis/admin.hydra.io/v1alpha1/models/${req["modelId"]}/servings?${fm.renderURLSearchParams(req, ["modelId"])}`, {...initReq, method: "GET"})
   }
+  static SortModels(req: SortModelsRequest, initReq?: fm.InitReq): Promise<GoogleProtobufEmpty.Empty> {
+    return fm.fetchReq<SortModelsRequest, GoogleProtobufEmpty.Empty>(`/apis/admin.hydra.io/v1alpha1/models:sort`, {...initReq, method: "POST", body: JSON.stringify(req, fm.replacer)})
+  }
 }
 export class AdminModelTemplateManagement {
   static CreateModelTemplate(req: HydraManagement_apiModelV1alpha1Model.CreateModelTemplateRequest, initReq?: fm.InitReq): Promise<HydraManagement_apiModelV1alpha1Model.ModelTemplate> {

package/admin-api/model/v1alpha1/provider.pb.ts

@@ -6,6 +6,13 @@
 
 import * as HydraCommonCommon from "../../../common/common.pb"
 import * as fm from "../../../fetch.pb"
+import * as GoogleProtobufTimestamp from "../../../google/protobuf/timestamp.pb"
+
+export enum DuplicatePolicy {
+  DUPLICATE_POLICY_SKIP = "DUPLICATE_POLICY_SKIP",
+  DUPLICATE_POLICY_UPDATE = "DUPLICATE_POLICY_UPDATE",
+}
+
 export type CreateProviderRequest = {
   providerId?: string
   providerName?: HydraCommonCommon.I18nName
@@ -18,6 +25,7 @@ export type Provider = {
   providerName?: HydraCommonCommon.I18nName
   providerDescription?: HydraCommonCommon.I18nName
   providerAvatar?: string
+  createTime?: GoogleProtobufTimestamp.Timestamp
 }
 
 export type UpdateProviderRequest = {
@@ -40,6 +48,30 @@ export type SingleProviderRequest = {
   providerId?: string
 }
 
+export type BatchCreateProvidersRequest = {
+  providers?: CreateProviderRequest[]
+  duplicatePolicy?: DuplicatePolicy
+}
+
+export type BatchCreateProvidersResponse = {
+  items?: Provider[]
+  skippedNames?: string[]
+}
+
+export type ParsedProvider = {
+  provider?: Provider
+  exist?: boolean
+  error?: string
+}
+
+export type ParseProvidersRequest = {
+  providers?: CreateProviderRequest[]
+}
+
+export type ParseProvidersResponse = {
+  items?: ParsedProvider[]
+}
+
 export class AdminProviderManagement {
   static CreateProvider(req: CreateProviderRequest, initReq?: fm.InitReq): Promise<Provider> {
     return fm.fetchReq<CreateProviderRequest, Provider>(`/apis/admin.hydra.io/v1alpha1/providers`, {...initReq, method: "POST", body: JSON.stringify(req, fm.replacer)})
@@ -56,4 +88,10 @@ export class AdminProviderManagement {
   static DeleteProvider(req: SingleProviderRequest, initReq?: fm.InitReq): Promise<Provider> {
     return fm.fetchReq<SingleProviderRequest, Provider>(`/apis/admin.hydra.io/v1alpha1/providers/${req["providerId"]}`, {...initReq, method: "DELETE"})
   }
+  static ParseProviderSpecs(req: ParseProvidersRequest, initReq?: fm.InitReq): Promise<ParseProvidersResponse> {
+    return fm.fetchReq<ParseProvidersRequest, ParseProvidersResponse>(`/apis/admin.hydra.io/v1alpha1/providers:parse`, {...initReq, method: "POST", body: JSON.stringify(req, fm.replacer)})
+  }
+  static BatchCreateProviders(req: BatchCreateProvidersRequest, initReq?: fm.InitReq): Promise<BatchCreateProvidersResponse> {
+    return fm.fetchReq<BatchCreateProvidersRequest, BatchCreateProvidersResponse>(`/apis/admin.hydra.io/v1alpha1/providers:batch_create`, {...initReq, method: "POST", body: JSON.stringify(req, fm.replacer)})
+  }
 }

package/common/errorcode.ts

@@ -30,6 +30,9 @@ export enum ErrorCode {
   // EN: Request mode not match
   // ZH: 请求模式不匹配。
   SYSTEM_REQUEST_MODE_ERROR = "SYSTEM-REQUEST_MODE_ERROR",
+  // EN: upstream service unavailable: %v
+  // ZH: 上游服务暂时不可用：%v
+  SYSTEM_UPSTREAM_UNAVAILABLE = "SYSTEM-UPSTREAM_UNAVAILABLE",
   // EN: Requested resource conflicted error: %v
   // ZH: 请求的资源冲突错误：%v
   USER_CONFLICT_ERROR = "USER-CONFLICT_ERROR",

package/management-api/apikey/v1alpha1/apikey.pb.ts

@@ -85,6 +85,16 @@ export type DeleteWSAPIKeyRequest = {
   workspace?: number
 }
 
+export type GetWSAPIKeySecretRequest = {
+  id?: string
+  workspace?: number
+}
+
+export type APIKeySecret = {
+  id?: string
+  key?: string
+}
+
 export type UpdateWSAPIKeyQuotaRequest = {
   id?: string
   workspace?: number
@@ -102,6 +112,10 @@ export type DeleteAPIKeyRequest = {
   id?: string
 }
 
+export type GetAPIKeySecretRequest = {
+  id?: string
+}
+
 export type CheckAPIKeyCreationRequest = {
 }
 
@@ -157,6 +171,12 @@ export class APIKeyManagement {
   static ListAPIKey(req: ListAPIKeyRequest, initReq?: fm.InitReq): Promise<ListAPIKeyResponse> {
     return fm.fetchReq<ListAPIKeyRequest, ListAPIKeyResponse>(`/apis/hydra.io/v1alpha1/apikeys?${fm.renderURLSearchParams(req, [])}`, {...initReq, method: "GET"})
   }
+  static ListAPIKeyMasked(req: ListAPIKeyRequest, initReq?: fm.InitReq): Promise<ListAPIKeyResponse> {
+    return fm.fetchReq<ListAPIKeyRequest, ListAPIKeyResponse>(`/apis/hydra.io/v1alpha1/masked-apikeys?${fm.renderURLSearchParams(req, [])}`, {...initReq, method: "GET"})
+  }
+  static GetAPIKeySecret(req: GetAPIKeySecretRequest, initReq?: fm.InitReq): Promise<APIKeySecret> {
+    return fm.fetchReq<GetAPIKeySecretRequest, APIKeySecret>(`/apis/hydra.io/v1alpha1/apikeys/${req["id"]}/secret?${fm.renderURLSearchParams(req, ["id"])}`, {...initReq, method: "GET"})
+  }
   static DeleteAPIKey(req: DeleteAPIKeyRequest, initReq?: fm.InitReq): Promise<APIKey> {
     return fm.fetchReq<DeleteAPIKeyRequest, APIKey>(`/apis/hydra.io/v1alpha1/apikeys/${req["id"]}`, {...initReq, method: "DELETE", body: JSON.stringify(req, fm.replacer)})
   }
@@ -177,6 +197,12 @@ export class WSAPIKeyManagement {
   static ListMyWSAPIKeys(req: ListWSAPIKeyRequest, initReq?: fm.InitReq): Promise<ListAPIKeyResponse> {
     return fm.fetchReq<ListWSAPIKeyRequest, ListAPIKeyResponse>(`/apis/hydra.io/v1alpha1/workspaces/${req["workspace"]}/my-apikeys?${fm.renderURLSearchParams(req, ["workspace"])}`, {...initReq, method: "GET"})
   }
+  static ListWSAPIKeyMasked(req: ListWSAPIKeyRequest, initReq?: fm.InitReq): Promise<ListAPIKeyResponse> {
+    return fm.fetchReq<ListWSAPIKeyRequest, ListAPIKeyResponse>(`/apis/hydra.io/v1alpha1/workspaces/${req["workspace"]}/masked-apikeys?${fm.renderURLSearchParams(req, ["workspace"])}`, {...initReq, method: "GET"})
+  }
+  static GetWSAPIKeySecret(req: GetWSAPIKeySecretRequest, initReq?: fm.InitReq): Promise<APIKeySecret> {
+    return fm.fetchReq<GetWSAPIKeySecretRequest, APIKeySecret>(`/apis/hydra.io/v1alpha1/workspaces/${req["workspace"]}/apikeys/${req["id"]}/secret?${fm.renderURLSearchParams(req, ["workspace", "id"])}`, {...initReq, method: "GET"})
+  }
   static DeleteWSAPIKey(req: DeleteWSAPIKeyRequest, initReq?: fm.InitReq): Promise<APIKey> {
     return fm.fetchReq<DeleteWSAPIKeyRequest, APIKey>(`/apis/hydra.io/v1alpha1/workspaces/${req["workspace"]}/apikeys/${req["id"]}`, {...initReq, method: "DELETE", body: JSON.stringify(req, fm.replacer)})
   }

package/management-api/core/v1alpha1/management.pb.ts

@@ -28,6 +28,8 @@ export type InferenceRuntime = {
   name?: HydraCommonCommon.I18nName
   runtimeCommand?: string
   distributedSupported?: boolean
+  leaderCommand?: string
+  workerCommand?: string
 }
 
 export type ListGPUTypesRequest = {

package/management-api/core/v1alpha1/permissions.pb.ts

@@ -18,6 +18,7 @@ export enum UserPermission {
   APIKEY_GET = "APIKEY_GET",
   APIKEY_CREATE = "APIKEY_CREATE",
   APIKEY_DELETE = "APIKEY_DELETE",
+  APIKEY_UPDATE = "APIKEY_UPDATE",
   CLUSTER_GET = "CLUSTER_GET",
   FILES_STORAGE_GET = "FILES_STORAGE_GET",
   FILES_STORAGE_CREATE = "FILES_STORAGE_CREATE",

package/management-api/dashboard/v1alpha1/dashboard.pb.ts

@@ -70,7 +70,7 @@ export type ListWSInstanceTokenUsageRequest = {
 }
 
 export type InstanceTokenUsage = {
-  instanceId?: string
+  apiKeyId?: string
   instanceName?: string
   modelName?: string
   inputTokens?: string
@@ -79,6 +79,7 @@ export type InstanceTokenUsage = {
   totalTokens?: string
   requestCount?: string
   lastUsedTime?: GoogleProtobufTimestamp.Timestamp
+  username?: string
 }
 
 export type ListInstanceTokenUsageResponse = {

package/management-api/model/v1alpha1/model.pb.ts

@@ -9,6 +9,15 @@ import * as fm from "../../../fetch.pb"
 import * as GoogleProtobufEmpty from "../../../google/protobuf/empty.pb"
 import * as GoogleProtobufTimestamp from "../../../google/protobuf/timestamp.pb"
 
+type Absent<T, K extends keyof T> = { [k in Exclude<keyof T, K>]?: undefined };
+type OneOf<T> =
+  | { [k in keyof T]?: undefined }
+  | (
+    keyof T extends infer K ?
+      (K extends string & keyof T ? { [k in K]: T[K] } & Absent<T, K>
+        : never)
+    : never);
+
 export enum ModelSupportFeature {
   MODEL_SUPPORT_TYPE_UNSPECIFIED = "MODEL_SUPPORT_TYPE_UNSPECIFIED",
   TEXT_GENERATION = "TEXT_GENERATION",
@@ -22,6 +31,22 @@ export enum ModelSupportFeature {
   AUDIO_TO_AUDIO = "AUDIO_TO_AUDIO",
   AUDIO_TO_TEXT = "AUDIO_TO_TEXT",
   TOOLS = "TOOLS",
+  TEXT_TO_AUDIO = "TEXT_TO_AUDIO",
+}
+
+export enum ContextLengthRange {
+  CONTEXT_LENGTH_RANGE_UNSPECIFIED = "CONTEXT_LENGTH_RANGE_UNSPECIFIED",
+  CTX_0_32K = "CTX_0_32K",
+  CTX_32K_128K = "CTX_32K_128K",
+  CTX_128K_256K = "CTX_128K_256K",
+  CTX_256K_1M = "CTX_256K_1M",
+}
+
+export enum ModelOperationTag {
+  MODEL_OPERATION_TAG_UNSPECIFIED = "MODEL_OPERATION_TAG_UNSPECIFIED",
+  FEATURED = "FEATURED",
+  HOT = "HOT",
+  DEPRECATING = "DEPRECATING",
 }
 
 export enum SkuDiscountScheduleType {
@@ -43,6 +68,16 @@ export enum DeployTemplateInferenceRuntime {
   IMAGE_GEN = "IMAGE_GEN",
 }
 
+export type ModelSupportFeatureRoute = {
+  modelSupportFeature?: ModelSupportFeature
+  defaultPath?: string
+  dependency?: ModelSupportFeature
+}
+
+export type ModelSupportFeatureConfig = {
+  items?: ModelSupportFeatureRoute[]
+}
+
 export type ModelPriceDiscountInfo = {
   discountRate?: number
   discountName?: string
@@ -89,6 +124,10 @@ export type Model = {
   publicModelConfig?: PublicModelConfig
   modelDeploymentsExists?: boolean
   visible?: string
+  parameterSize?: number
+  contextLength?: ContextLengthRange
+  operationTags?: ModelOperationTag[]
+  publishTimestamp?: GoogleProtobufTimestamp.Timestamp
 }
 
 export type ListModelRequest = {
@@ -138,6 +177,9 @@ export type DeployTemplate = {
   runtimeCommand?: string
   templateName?: string
   templateDescription?: string
+  workerCommand?: string
+  workerEnv?: EnvVar[]
+  distributed?: boolean
 }
 
 export type EnvVar = {
@@ -188,6 +230,10 @@ export type WSModel = {
   modelDeploymentsExists?: boolean
   visible?: string
   tokenWeight?: string
+  parameterSize?: number
+  contextLength?: ContextLengthRange
+  operationTags?: ModelOperationTag[]
+  publishTimestamp?: GoogleProtobufTimestamp.Timestamp
 }
 
 export type CreatePrivateModelRequest = {
@@ -241,18 +287,29 @@ export type DeleteModelTemplateRequest = {
   modelId?: string
 }
 
-export type ListModelTemplatesRequest = {
+
+/* hydra modified */ export type BaseListModelTemplatesRequest = {
   workspace?: number
   modelId?: string
   inferenceRuntime?: string
   page?: HydraCommonCommon.Pagination
 }
 
+export type ListModelTemplatesRequest = BaseListModelTemplatesRequest
+  & OneOf<{ distributed: boolean }>
+
 export type ListModelTemplatesResponse = {
   items?: ModelTemplate[]
   page?: HydraCommonCommon.Pagination
 }
 
+export type GetModelSupportFeatureConfigRequest = {
+}
+
+export type GetModelSupportFeatureConfigResponse = {
+  config?: ModelSupportFeatureConfig
+}
+
 export class ModelManagement {
   static GetModel(req: SingleModelRequest, initReq?: fm.InitReq): Promise<Model> {
     return fm.fetchReq<SingleModelRequest, Model>(`/apis/hydra.io/v1alpha1/models/${req["modelId"]}?${fm.renderURLSearchParams(req, ["modelId"])}`, {...initReq, method: "GET"})
@@ -300,4 +357,9 @@ export class WSModelTemplateManagement {
   static ListWSModelTemplates(req: ListModelTemplatesRequest, initReq?: fm.InitReq): Promise<ListModelTemplatesResponse> {
     return fm.fetchReq<ListModelTemplatesRequest, ListModelTemplatesResponse>(`/apis/hydra.io/v1alpha1/workspaces/${req["workspace"]}/models/${req["modelId"]}/model-templates?${fm.renderURLSearchParams(req, ["workspace", "modelId"])}`, {...initReq, method: "GET"})
   }
+}
+export class ModelConfigManagement {
+  static GetModelSupportFeatureConfig(req: GetModelSupportFeatureConfigRequest, initReq?: fm.InitReq): Promise<GetModelSupportFeatureConfigResponse> {
+    return fm.fetchReq<GetModelSupportFeatureConfigRequest, GetModelSupportFeatureConfigResponse>(`/apis/hydra.io/v1alpha1/model-support-features/config?${fm.renderURLSearchParams(req, [])}`, {...initReq, method: "GET"})
+  }
 }

package/management-api/model_serving/v1alpha1/model_serving.pb.ts

@@ -65,9 +65,17 @@ export enum DeployDetectConditionType {
   MODEL_IMAGE_DOWNLOADED = "MODEL_IMAGE_DOWNLOADED",
 }
 
+export enum BuiltinEnvVarScope {
+  SCOPE_ALL = "SCOPE_ALL",
+  SCOPE_DISTRIBUTED = "SCOPE_DISTRIBUTED",
+  SCOPE_LEADER = "SCOPE_LEADER",
+  SCOPE_WORKER = "SCOPE_WORKER",
+}
+
 export enum ModelServingActionRequestAction {
   MODEL_SERVING_ACTION_UNSPECIFIED = "MODEL_SERVING_ACTION_UNSPECIFIED",
   START = "START",
+  STOP = "STOP",
 }
 
 export enum WSModelServingActionRequestAction {
@@ -156,6 +164,8 @@ export type RuntimeConfig = {
   imagePullSecretName?: string[]
   runtimeCommand?: string
   env?: EnvVar[]
+  workerCommand?: string
+  workerEnv?: EnvVar[]
 }
 
 export type CreateModelServingRequest = {
@@ -343,12 +353,14 @@ export type ListBuiltinEnvVarsRequest = {
 
 export type BuiltinEnvVar = {
   name?: string
-  description?: string
+  description?: HydraCommonCommon.I18nName
   example?: string
 }
 
 export type ListBuiltinEnvVarsResponse = {
-  items?: BuiltinEnvVar[]
+  single?: BuiltinEnvVar[]
+  leader?: BuiltinEnvVar[]
+  worker?: BuiltinEnvVar[]
 }
 
 export class ModelServingManagement {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@daocloud-proto/hydra",
-  "version": "v0.16.0",
+  "version": "v0.17.0-rc1-dev-1",
   "description": "",
   "author": "",
   "license": "ISC"