@daocloud-proto/hydra 0.16.0-dev-7 → 0.16.0-dev-9

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

@@ -0,0 +1,337 @@
+/* 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_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 = {
+  ingressReady?: boolean
+  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
+  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/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@daocloud-proto/hydra",
-  "version": "v0.16.0-dev-7",
+  "version": "v0.16.0-dev-9",
   "description": "",
   "author": "",
   "license": "ISC"