@daocloud-proto/hydra 0.16.0-dev-10 → 0.16.0-dev-12

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

@@ -28,6 +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",
@@ -53,6 +59,7 @@ export type CreateMAASModelRequest = {
   loadBalanceConfig?: LoadBalanceConfig
   tokenWeight?: string
   workspaceVisibility?: MAASWorkspaceVisibility
+  capabilityRoutes?: MAASCapabilityRoute[]
 }
 
 export type MAASModel = {
@@ -66,6 +73,7 @@ export type MAASModel = {
   enabled?: boolean
   tokenWeight?: string
   workspaceVisibility?: MAASWorkspaceVisibility
+  capabilityRoutes?: MAASCapabilityRoute[]
 }
 
 export type LoadBalanceConfig = {
@@ -95,6 +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
@@ -118,6 +139,7 @@ export type UpdateMAASModelRequest = {
   upstreamModels?: UpstreamModel[]
   tokenWeight?: string
   workspaceVisibility?: MAASWorkspaceVisibility
+  capabilityRoutes?: MAASCapabilityRoute[]
 }
 
 export type UpdateMAASModelStatusRequest = {

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",
@@ -148,6 +157,9 @@ export type DeployTemplate = {
   runtimeCommand?: string
   templateName?: string
   templateDescription?: string
+  workerCommand?: string
+  workerEnv?: EnvVar[]
+  distributed?: boolean
 }
 
 export type EnvVar = {
@@ -251,13 +263,17 @@ 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

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

@@ -65,6 +65,13 @@ 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",
@@ -157,6 +164,8 @@ export type RuntimeConfig = {
   imagePullSecretName?: string[]
   runtimeCommand?: string
   env?: EnvVar[]
+  workerCommand?: string
+  workerEnv?: EnvVar[]
 }
 
 export type CreateModelServingRequest = {
@@ -349,7 +358,9 @@ export type BuiltinEnvVar = {
 }
 
 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-dev-10",
+  "version": "v0.16.0-dev-12",
   "description": "",
   "author": "",
   "license": "ISC"