fix: model reload state - reduce model unload events emit

core/src/browser/models/manager.ts

@@ -18,7 +18,14 @@ export class ModelManager {
    * @param model - The model to register.
    */
   register<T extends Model>(model: T) {
+    if (this.models.has(model.id)) {
+      this.models.set(model.id, {
+        ...model,
+        ...this.models.get(model.id),
+      })
+    } else {
       this.models.set(model.id, model)
+    }
     events.emit(ModelEvent.OnModelsUpdate, {})
   }
 

docs/src/pages/docs/built-in/llama-cpp.mdx

@@ -102,7 +102,7 @@ Enable the GPU acceleration option within the Jan application by following the [
     ],
     "size": 669000000
   },
-  "engine": "llama-cpp"
+  "engine": "nitro"
 }
 ```
 ### Step 2: Modify the `model.json`

extensions/inference-cortex-extension/package.json

@@ -51,6 +51,7 @@
     "decompress": "^4.2.1",
     "fetch-retry": "^5.0.6",
     "ky": "^1.7.2",
+    "p-queue": "^8.0.1",
     "rxjs": "^7.8.1",
     "tcp-port-used": "^1.0.2",
     "terminate": "2.6.1",

extensions/inference-cortex-extension/rollup.config.ts

@@ -114,7 +114,7 @@ export default [
         ]),
         NODE: JSON.stringify(`${packageJson.name}/${packageJson.node}`),
         DEFAULT_SETTINGS: JSON.stringify(defaultSettingJson),
-        CORTEX_API_URL: JSON.stringify('http://127.0.0.1:39291/v1'),
+        CORTEX_API_URL: JSON.stringify('http://127.0.0.1:39291'),
       }),
       // Allow json resolution
       json(),

extensions/inference-cortex-extension/src/index.ts

@@ -16,7 +16,7 @@ import {
   LocalOAIEngine,
   InferenceEngine,
 } from '@janhq/core'
-
+import PQueue from 'p-queue'
 import ky from 'ky'
 
 /**
@@ -28,12 +28,14 @@ export default class JanInferenceCortexExtension extends LocalOAIEngine {
   // DEPRECATED
   nodeModule: string = 'node'
 
+  queue = new PQueue({ concurrency: 1 })
+
   provider: string = InferenceEngine.cortex
 
   /**
    * The URL for making inference requests.
    */
-  inferenceUrl = `${CORTEX_API_URL}/chat/completions`
+  inferenceUrl = `${CORTEX_API_URL}/v1/chat/completions`
 
   /**
    * Subscribes to events emitted by the @janhq/core package.
@@ -47,7 +49,9 @@ export default class JanInferenceCortexExtension extends LocalOAIEngine {
 
     // Run the process watchdog
     const systemInfo = await systemInformation()
-    executeOnMain(NODE, 'run', systemInfo)
+    await executeOnMain(NODE, 'run', systemInfo)
+
+    this.queue.add(() => this.healthz())
   }
 
   onUnload(): void {
@@ -61,16 +65,19 @@ export default class JanInferenceCortexExtension extends LocalOAIEngine {
     // Legacy model cache - should import
     if (model.engine === InferenceEngine.nitro && model.file_path) {
       // Try importing the model
-      await ky
+      const modelPath = await this.modelPath(model)
-        .post(`${CORTEX_API_URL}/models/${model.id}`, {
+      await this.queue.add(() =>
-          json: { model: model.id, modelPath: await this.modelPath(model) },
+        ky
+          .post(`${CORTEX_API_URL}/v1/models/${model.id}`, {
+            json: { model: model.id, modelPath: modelPath },
           })
           .json()
           .catch((e) => log(e.message ?? e ?? ''))
+      )
     }
 
-    return ky
+    return await ky
-      .post(`${CORTEX_API_URL}/models/start`, {
+      .post(`${CORTEX_API_URL}/v1/models/start`, {
         json: {
           ...model.settings,
           model: model.id,
@@ -89,7 +96,7 @@ export default class JanInferenceCortexExtension extends LocalOAIEngine {
 
   override async unloadModel(model: Model): Promise<void> {
     return ky
-      .post(`${CORTEX_API_URL}/models/stop`, {
+      .post(`${CORTEX_API_URL}/v1/models/stop`, {
         json: { model: model.id },
       })
       .json()
@@ -108,4 +115,19 @@ export default class JanInferenceCortexExtension extends LocalOAIEngine {
         model.id,
     ])
   }
+
+  /**
+   * Do health check on cortex.cpp
+   * @returns
+   */
+  healthz(): Promise<void> {
+    return ky
+      .get(`${CORTEX_API_URL}/healthz`, {
+        retry: {
+          limit: 10,
+          methods: ['get'],
+        },
+      })
+      .then(() => {})
+  }
 }

extensions/inference-cortex-extension/tsconfig.json

@@ -1,9 +1,8 @@
 {
   "compilerOptions": {
     "moduleResolution": "node",
-    "target": "ES2015",
+    "target": "es2016",
-    "module": "ES2020",
+    "module": "esnext",
-    "lib": ["es2015", "es2016", "es2017", "dom"],
     "strict": true,
     "sourceMap": true,
     "declaration": true,

web/containers/Loader/ModelReload.tsx

@@ -44,6 +44,11 @@ export default function ModelReload() {
           Reloading model {stateModel.model?.id}
         </span>
       </div>
+      <div className="my-4 mb-2 text-center">
+        <span className="text-[hsla(var(--text-secondary)]">
+          Model is reloading to apply new changes.
+        </span>
+      </div>
     </div>
   )
 }

web/hooks/useActiveModel.ts

@@ -51,6 +51,10 @@ export function useActiveModel() {
       console.debug(`Model ${modelId} is already initialized. Ignore..`)
       return Promise.resolve()
     }
+
+    if (activeModel) {
+      stopModel(activeModel)
+    }
     setPendingModelLoad(true)
 
     let model = downloadedModelsRef?.current.find((e) => e.id === modelId)
@@ -113,7 +117,7 @@ export function useActiveModel() {
         setStateModel(() => ({
           state: 'start',
           loading: false,
-          model,
+          undefined,
         }))
 
         if (!pendingModelLoad && abortable) {
@@ -130,12 +134,12 @@ export function useActiveModel() {
       })
   }
 
-  const stopModel = useCallback(async () => {
+  const stopModel = useCallback(
-    const stoppingModel = activeModel || stateModel.model
+    async (model?: Model) => {
+      const stoppingModel = model ?? activeModel ?? stateModel.model
       if (!stoppingModel || (stateModel.state === 'stop' && stateModel.loading))
         return
 
-    setStateModel({ state: 'stop', loading: true, model: stoppingModel })
       const engine = EngineManager.instance().get(stoppingModel.engine)
       return engine
         ?.unloadModel(stoppingModel)
@@ -145,13 +149,15 @@ export function useActiveModel() {
           setStateModel({ state: 'start', loading: false, model: undefined })
           setPendingModelLoad(false)
         })
-  }, [
+    },
+    [
       activeModel,
-    setActiveModel,
       setStateModel,
+      setActiveModel,
       setPendingModelLoad,
       stateModel,
-  ])
+    ]
+  )
 
   const stopInference = useCallback(async () => {
     // Loading model

web/hooks/useModels.ts

@@ -31,10 +31,9 @@ const useModels = () => {
   const getData = useCallback(() => {
     const getDownloadedModels = async () => {
       const localModels = await getModels()
-      const remoteModels = ModelManager.instance()
+      const hubModels = ModelManager.instance().models.values().toArray()
-        .models.values()
+
-        .toArray()
+      const remoteModels = hubModels.filter((e) => !isLocalEngine(e.engine))
-        .filter((e) => !isLocalEngine(e.engine))
       setDownloadedModels([...localModels, ...remoteModels])
     }
 

web/screens/Thread/ThreadCenterPanel/index.tsx

@@ -199,16 +199,7 @@ const ThreadCenterPanel = () => {
 
           {!engineParamsUpdate && <ModelStart />}
 
-          {reloadModel && (
+          {reloadModel && <ModelReload />}
-            <Fragment>
-              <ModelReload />
-              <div className="mb-2 text-center">
-                <span className="text-[hsla(var(--text-secondary)]">
-                  Model is reloading to apply new changes.
-                </span>
-              </div>
-            </Fragment>
-          )}
 
           {activeModel && isGeneratingResponse && <GenerateResponse />}
           <ChatInput />

web/screens/Thread/ThreadRightPanel/index.tsx

@@ -15,6 +15,8 @@ import {
 
 import { useAtom, useAtomValue, useSetAtom } from 'jotai'
 
+import { useDebouncedCallback } from 'use-debounce'
+
 import CopyOverInstruction from '@/containers/CopyInstruction'
 import EngineSetting from '@/containers/EngineSetting'
 import ModelDropdown from '@/containers/ModelDropdown'
@@ -168,6 +170,10 @@ const ThreadRightPanel = () => {
     [activeThread, updateThreadMetadata]
   )
 
+  const resetModel = useDebouncedCallback(() => {
+    stopModel()
+  }, 300)
+
   const onValueChanged = useCallback(
     (key: string, value: string | number | boolean) => {
       if (!activeThread) {
@@ -175,7 +181,7 @@ const ThreadRightPanel = () => {
       }
 
       setEngineParamsUpdate(true)
-      stopModel()
+      resetModel()
 
       updateModelParameter(activeThread, {
         params: { [key]: value },
@@ -207,7 +213,7 @@ const ThreadRightPanel = () => {
         }
       }
     },
-    [activeThread, setEngineParamsUpdate, stopModel, updateModelParameter]
+    [activeThread, resetModel, setEngineParamsUpdate, updateModelParameter]
   )
 
   if (!activeThread) {