Merge pull request #4049 from janhq/fix/openai-o1-model-parameters

fix: correct OpenAI o1 model parameters

core/src/browser/extensions/engines/helpers/sse.ts

@@ -45,7 +45,9 @@ export function requestInference(
           subscriber.complete()
           return
         }
-        if (model.parameters?.stream === false) {
+        // There could be overriden stream parameter in the model
+        // that is set in request body (transformed payload)
+        if (requestBody?.stream === false || model.parameters?.stream === false) {
           const data = await response.json()
           if (transformResponse) {
             subscriber.next(transformResponse(data))

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

@@ -69,12 +69,13 @@ export default class JanInferenceCortexExtension extends LocalOAIEngine {
 
     super.onLoad()
 
-    await this.queue.add(() => this.clean())
-    this.queue.add(() => this.healthz())
-    this.queue.add(() => this.setDefaultEngine(systemInfo))
+    this.queue.add(() => this.clean())
+    
     // Run the process watchdog
     const systemInfo = await systemInformation()
-    await executeOnMain(NODE, 'run', systemInfo)
+    this.queue.add(() => executeOnMain(NODE, 'run', systemInfo))
+    this.queue.add(() => this.healthz())
+    this.queue.add(() => this.setDefaultEngine(systemInfo))
     this.subscribeToEvents()
 
     window.addEventListener('beforeunload', () => {

extensions/inference-openai-extension/resources/models.json

@@ -97,11 +97,9 @@
     "format": "api",
     "settings": {},
     "parameters": {
-      "max_tokens": 4096,
-      "temperature": 0.7,
-      "top_p": 0.95,
-      "stream": true,
-      "stop": [],
+      "temperature": 1,
+      "top_p": 1,
+      "max_tokens": 32768,
       "frequency_penalty": 0,
       "presence_penalty": 0
     },
@@ -125,11 +123,9 @@
     "format": "api",
     "settings": {},
     "parameters": {
-      "max_tokens": 4096,
-      "temperature": 0.7,
-      "top_p": 0.95,
-      "stream": true,
-      "stop": [],
+      "temperature": 1,
+      "top_p": 1,
+      "max_tokens": 65536,
       "frequency_penalty": 0,
       "presence_penalty": 0
     },

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

@@ -76,11 +76,11 @@ export default class JanInferenceOpenAIExtension extends RemoteOAIEngine {
   transformPayload = (payload: OpenAIPayloadType): OpenAIPayloadType => {
     // Transform the payload for preview models
     if (this.previewModels.includes(payload.model)) {
-      const { max_tokens, temperature, top_p, stop, ...params } = payload
+      const { max_tokens, stop, ...params } = payload
       return {
         ...params,
         max_completion_tokens: max_tokens,
-        stream: false // o1 only support stream = false
+        stream: false, // o1 only support stream = false
       }
     }
     // Pass through for non-preview models

extensions/model-extension/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@janhq/model-extension",
   "productName": "Model Management",
-  "version": "1.0.34",
+  "version": "1.0.35",
   "description": "Model Management Extension provides model exploration and seamless downloads",
   "main": "dist/index.js",
   "author": "Jan <service@jan.ai>",

extensions/model-extension/src/index.ts

@@ -143,7 +143,10 @@ export default class JanModelExtension extends ModelExtension {
      * There is no model to import
      * just return fetched models
      */
-    if (!toImportModels.length) return fetchedModels
+    if (!toImportModels.length)
+      return fetchedModels.concat(
+        legacyModels.filter((e) => e.settings?.vision_model)
+      )
 
     console.log('To import models:', toImportModels.length)
     /**