feat: add gpus_per_replicas

src/pages/llmodels/config/index.ts

@@ -194,6 +194,11 @@ export const ScheduleValueMap = {
   SpecificGPUType: 'specific_gpu_type'
 };
 
+export const gpusCountTypeMap = {
+  Auto: 'auto',
+  Custom: 'custom'
+};
+
 export const scheduleList = [
   {
     label: 'models.form.scheduletype.auto',
@@ -324,6 +329,7 @@ export const excludeFields = [
   'backend_version',
   'ollama_library_model_name',
   'scheduleType',
+  'gpusCountType',
   'placement_strategy',
   'backend',
   'gpu_selector',

src/pages/llmodels/config/types.ts

@@ -27,6 +27,7 @@ export interface ListItem {
   access_policy: 'public' | 'authed' | 'allowed_users';
   gpu_selector?: {
     gpu_ids: string[];
+    gpus_per_replica?: number;
   };
   worker_selector?: object;
 }
@@ -63,11 +64,13 @@ export interface FormData {
     gpu_ids?: string[];
     gpu_type?: string;
     gpu_count?: number;
+    gpus_per_replica?: number;
   };
   placement_strategy?: string;
   cpu_offloading?: boolean;
   worker_selector?: object;
   scheduleType?: string;
+  gpusCountType?: string;
   name: string;
   replicas: number;
   description: string;
@@ -191,6 +194,7 @@ export interface CatalogSpec {
   worker_selector: Record<string, any>;
   gpu_selector: {
     gpu_ids: string[];
+    gpus_per_replica: number;
   };
   backend: string;
   backend_version: string;
@@ -220,6 +224,7 @@ export interface EvaluateSpec {
   worker_selector?: Record<string, any>;
   gpu_selector?: {
     gpu_ids: string[];
+    gpus_per_replica: number;
   };
   backend?: string;
   backend_version?: string;

src/pages/llmodels/config/utils.ts

@@ -1,4 +1,5 @@
 import _ from 'lodash';
+import { gpusCountTypeMap } from '.';
 import { backendOptionsMap } from '../config/backend-parameters';
 import { FormData } from './types';
 
@@ -33,6 +34,11 @@ export const generateGPUSelector = (data: any, gpuOptions: any[]) => {
   };
 };
 
+/**
+ * before submit the form, generate the gpu_selector field
+ * @param data
+ * @returns
+ */
 export const generateGPUIds = (data: FormData) => {
   const gpu_ids = _.get(data, 'gpu_selector.gpu_ids', []);
   console.log('generateGPUIds', gpu_ids);
@@ -54,10 +60,17 @@ export const generateGPUIds = (data: FormData) => {
     },
     []
   );
-
+  if (gpusCountTypeMap.Auto === data.gpusCountType) {
+    return {
+      gpu_selector: {
+        gpu_ids: result || []
+      }
+    };
+  }
   return {
     gpu_selector: {
-      gpu_ids: result || []
+      gpu_ids: result || [],
+      gpus_per_replica: data.gpu_selector?.gpus_per_replica || null
     }
   };
 };