fix: support load model configurations

core/src/browser/extensions/engines/AIEngine.ts

@@ -231,7 +231,7 @@ export abstract class AIEngine extends BaseExtension {
   /**
    * Loads a model into memory
    */
-  abstract load(modelId: string): Promise<SessionInfo>
+  abstract load(modelId: string, settings?: any): Promise<SessionInfo>
 
   /**
    * Unloads a model from memory

web-app/src/containers/ModelSetting.tsx

@@ -87,8 +87,10 @@ export function ModelSetting({
         ...(params as unknown as object),
       })
 
-      // Call debounced stopModel after updating the model
-      debouncedStopModel(model.id)
+      // Call debounced stopModel only when updating ctx_len or ngl
+      if (key === 'ctx_len' || key === 'ngl') {
+        debouncedStopModel(model.id)
+      }
     }
   }
 
@@ -106,7 +108,9 @@ export function ModelSetting({
       </SheetTrigger>
       <SheetContent className="h-[calc(100%-8px)] top-1 right-1 rounded-e-md overflow-y-auto">
         <SheetHeader>
-          <SheetTitle>{t('common:modelSettings.title', { modelId: model.id })}</SheetTitle>
+          <SheetTitle>
+            {t('common:modelSettings.title', { modelId: model.id })}
+          </SheetTitle>
           <SheetDescription>
             {t('common:modelSettings.description')}
           </SheetDescription>

web-app/src/hooks/useChat.ts

@@ -261,14 +261,36 @@ export const useChat = () => {
           !abortController.signal.aborted &&
           activeProvider
         ) {
+          const modelConfig = activeProvider.models.find(
+            (m) => m.id === selectedModel?.id
+          )
+
+          const modelSettings = modelConfig?.settings
+            ? Object.fromEntries(
+                Object.entries(modelConfig.settings)
+                  .filter(
+                    ([key, value]) =>
+                      key !== 'ctx_len' &&
+                      key !== 'ngl' &&
+                      value.controller_props?.value !== undefined &&
+                      value.controller_props?.value !== null &&
+                      value.controller_props?.value !== ''
+                  )
+                  .map(([key, value]) => [key, value.controller_props?.value])
+              )
+            : undefined
+
           const completion = await sendCompletion(
             activeThread,
             activeProvider,
             builder.getMessages(),
             abortController,
             availableTools,
             currentAssistant.parameters?.stream === false ? false : true,
-            currentAssistant.parameters as unknown as Record<string, object>
+            {
+              ...modelSettings,
+              ...currentAssistant.parameters,
+            } as unknown as Record<string, object>
           )
 
           if (!completion) throw new Error('No completion received')

web-app/src/services/__tests__/models.test.ts

@@ -1,4 +1,5 @@
 import { describe, it, expect, vi, beforeEach } from 'vitest'
+
 import {
   fetchModels,
   fetchModelCatalog,
@@ -10,9 +11,8 @@
   stopModel,
   stopAllModels,
   startModel,
-  configurePullOptions,
 } from '../models'
-import { EngineManager } from '@janhq/core'
+import { EngineManager, Model } from '@janhq/core'
 
 // Mock EngineManager
 vi.mock('@janhq/core', () => ({
@@ -118,7 +118,7 @@
         settings: [{ key: 'temperature', value: 0.7 }],
       }
 
-      await updateModel(model)
+      await updateModel(model as any)
 
       expect(mockEngine.updateSettings).toHaveBeenCalledWith(model.settings)
     })
@@ -209,7 +209,7 @@
 
   describe('startModel', () => {
     it('should start model successfully', async () => {
-      const provider = { provider: 'openai', models: [] } as ProviderObject
+      const provider = { provider: 'openai', models: [] } as any
       const model = 'model1'
       const mockSession = { id: 'session1' }
 
@@ -221,11 +221,11 @@
       const result = await startModel(provider, model)
 
       expect(result).toEqual(mockSession)
       expect(mockEngine.load).toHaveBeenCalledWith(model)
     })
 
     it('should handle start model error', async () => {
-      const provider = { provider: 'openai', models: [] } as ProviderObject
+      const provider = { provider: 'openai', models: [] } as any
       const model = 'model1'
       const error = new Error('Failed to start model')
 
@@ -237,7 +237,7 @@
       await expect(startModel(provider, model)).rejects.toThrow(error)
     })
     it('should not load model again', async () => {
-      const provider = { provider: 'openai', models: [] } as ProviderObject
+      const provider = { provider: 'openai', models: [] } as any
       const model = 'model1'
 
       mockEngine.getLoadedModels.mockResolvedValue({

web-app/src/services/models.ts

@@ -150,7 +150,29 @@ export const startModel = async (
   if (!engine) return undefined
 
   if ((await engine.getLoadedModels()).includes(model)) return undefined
-  return engine.load(model).catch((error) => {
+
+  // Find the model configuration to get settings
+  const modelConfig = provider.models.find((m) => m.id === model)
+
+  // Key mapping function to transform setting keys
+  const mapSettingKey = (key: string): string => {
+    const keyMappings: Record<string, string> = {
+      ctx_len: 'ctx_size',
+      ngl: 'n_gpu_layers',
+    }
+    return keyMappings[key] || key
+  }
+
+  const settings = modelConfig?.settings
+    ? Object.fromEntries(
+        Object.entries(modelConfig.settings).map(([key, value]) => [
+          mapSettingKey(key),
+          value.controller_props?.value,
+        ])
+      )
+    : undefined
+
+  return engine.load(model, settings).catch((error) => {
     console.error(
       `Failed to start model ${model} for provider ${provider.provider}:`,
       error