feat: gpt-3.5-turbo-instruct support, refactor: try fetching models if OpenRouter is set (#981)

* refactor: try fetching if OpenRouter api key is set

* feat: gpt-3.5-turbo-instruct support

* fix: use new assignment in getTokenizer

api/app/clients/ChatGPTClient.js

@@ -153,6 +153,11 @@ class ChatGPTClient extends BaseClient {
     } else {
       modelOptions.prompt = input;
     }
+
+    if (this.useOpenRouter && modelOptions.prompt) {
+      delete modelOptions.stop;
+    }
+
     const { debug } = this.options;
     const url = this.completionsUrl;
     if (debug) {

api/app/clients/OpenAIClient.js

@@ -73,21 +73,22 @@ class OpenAIClient extends BaseClient {
       this.useOpenRouter = true;
     }
 
+    const { model } = this.modelOptions;
+
     this.isChatCompletion =
       this.useOpenRouter ||
       this.options.reverseProxyUrl ||
       this.options.localAI ||
-      this.modelOptions.model.startsWith('gpt-');
+      model.includes('gpt-');
     this.isChatGptModel = this.isChatCompletion;
-    if (this.modelOptions.model === 'text-davinci-003') {
+    if (model.includes('text-davinci-003') || model.includes('instruct')) {
       this.isChatCompletion = false;
       this.isChatGptModel = false;
     }
     const { isChatGptModel } = this;
     this.isUnofficialChatGptModel =
-      this.modelOptions.model.startsWith('text-chat') ||
+      model.startsWith('text-chat') || model.startsWith('text-davinci-002-render');
-      this.modelOptions.model.startsWith('text-davinci-002-render');
+    this.maxContextTokens = maxTokensMap[model] ?? 4095; // 1 less than maximum
-    this.maxContextTokens = maxTokensMap[this.modelOptions.model] ?? 4095; // 1 less than maximum
     this.maxResponseTokens = this.modelOptions.max_tokens || 1024;
     this.maxPromptTokens =
       this.options.maxPromptTokens || this.maxContextTokens - this.maxResponseTokens;
@@ -168,8 +169,9 @@ class OpenAIClient extends BaseClient {
       tokenizer = this.constructor.getTokenizer(this.encoding, true, extendSpecialTokens);
     } else {
       try {
-        this.encoding = this.modelOptions.model;
+        const { model } = this.modelOptions;
-        tokenizer = this.constructor.getTokenizer(this.modelOptions.model, true);
+        this.encoding = model.includes('instruct') ? 'text-davinci-003' : model;
+        tokenizer = this.constructor.getTokenizer(this.encoding, true);
       } catch {
         tokenizer = this.constructor.getTokenizer(this.encoding, true);
       }
@@ -354,6 +356,8 @@ class OpenAIClient extends BaseClient {
           if (this.isChatCompletion) {
             token =
               progressMessage.choices?.[0]?.delta?.content ?? progressMessage.choices?.[0]?.text;
+          } else {
+            token = progressMessage.choices?.[0]?.text;
           }
 
           if (!token && this.useOpenRouter) {

api/server/services/ModelService.js

@@ -88,12 +88,11 @@ const getOpenAIModels = async (opts = { azure: false, plugins: false }) => {
     return models;
   }
 
-  if (userProvidedOpenAI) {
+  if (userProvidedOpenAI && !OPENROUTER_API_KEY) {
     return models;
   }
 
-  models = await fetchOpenAIModels(opts, models);
+  return await fetchOpenAIModels(opts, models);
-  return models;
 };
 
 const getChatGPTBrowserModels = () => {