🪙 feat: Configure Max Context and Output Tokens (#2648)

* chore: make frequent 'error' log into 'debug' log

* feat: add maxContextTokens as a conversation field

* refactor(settings): increase popover height

* feat: add DynamicInputNumber and maxContextTokens to all endpoints that support it (frontend), fix schema

* feat: maxContextTokens handling (backend)

* style: revert popover height

* feat: max tokens

* fix: Ollama Vision firebase compatibility

* fix: Ollama Vision, use message_file_map to determine multimodal request

* refactor: bring back MobileNav and improve title styling

api/app/clients/AnthropicClient.js

@@ -75,7 +75,9 @@ class AnthropicClient extends BaseClient {
     this.options.attachments?.then((attachments) => this.checkVisionRequest(attachments));
 
     this.maxContextTokens =
-      getModelMaxTokens(this.modelOptions.model, EModelEndpoint.anthropic) ?? 100000;
+      this.options.maxContextTokens ??
+      getModelMaxTokens(this.modelOptions.model, EModelEndpoint.anthropic) ??
+      100000;
     this.maxResponseTokens = this.modelOptions.maxOutputTokens || 1500;
     this.maxPromptTokens =
       this.options.maxPromptTokens || this.maxContextTokens - this.maxResponseTokens;
@@ -652,6 +654,7 @@ class AnthropicClient extends BaseClient {
 
   getSaveOptions() {
     return {
+      maxContextTokens: this.options.maxContextTokens,
       promptPrefix: this.options.promptPrefix,
       modelLabel: this.options.modelLabel,
       resendFiles: this.options.resendFiles,

api/app/clients/GoogleClient.js

@@ -138,7 +138,10 @@ class GoogleClient extends BaseClient {
       !isGenerativeModel && !isChatModel && /code|text/.test(this.modelOptions.model);
     const { isTextModel } = this;
 
-    this.maxContextTokens = getModelMaxTokens(this.modelOptions.model, EModelEndpoint.google);
+    this.maxContextTokens =
+      this.options.maxContextTokens ??
+      getModelMaxTokens(this.modelOptions.model, EModelEndpoint.google);
+
     // The max prompt tokens is determined by the max context tokens minus the max response tokens.
     // Earlier messages will be dropped until the prompt is within the limit.
     this.maxResponseTokens = this.modelOptions.maxOutputTokens || settings.maxOutputTokens.default;

api/app/clients/OpenAIClient.js

@@ -161,11 +161,13 @@ class OpenAIClient extends BaseClient {
       model.startsWith('text-chat') || model.startsWith('text-davinci-002-render');
 
     this.maxContextTokens =
+      this.options.maxContextTokens ??
       getModelMaxTokens(
         model,
         this.options.endpointType ?? this.options.endpoint,
         this.options.endpointTokenConfig,
-      ) ?? 4095; // 1 less than maximum
+      ) ??
+      4095; // 1 less than maximum
 
     if (this.shouldSummarize) {
       this.maxContextTokens = Math.floor(this.maxContextTokens / 2);
@@ -407,6 +409,7 @@ class OpenAIClient extends BaseClient {
 
   getSaveOptions() {
     return {
+      maxContextTokens: this.options.maxContextTokens,
       chatGptLabel: this.options.chatGptLabel,
       promptPrefix: this.options.promptPrefix,
       resendFiles: this.options.resendFiles,
@@ -435,7 +438,11 @@ class OpenAIClient extends BaseClient {
    * @returns {Promise<MongoFile[]>}
    */
   async addImageURLs(message, attachments) {
-    const { files, image_urls } = await encodeAndFormat(this.options.req, attachments);
+    const { files, image_urls } = await encodeAndFormat(
+      this.options.req,
+      attachments,
+      this.options.endpoint,
+    );
     message.image_urls = image_urls.length ? image_urls : undefined;
     return files;
   }
@@ -1158,7 +1165,7 @@ ${convo}
         });
       }
 
-      if (this.options.attachments && this.options.endpoint?.toLowerCase() === 'ollama') {
+      if (this.message_file_map && this.options.endpoint?.toLowerCase() === 'ollama') {
         const ollamaClient = new OllamaClient({ baseURL });
         return await ollamaClient.chatCompletion({
           payload: modelOptions,

api/app/clients/specs/FakeClient.js

@@ -40,7 +40,8 @@ class FakeClient extends BaseClient {
       };
     }
 
-    this.maxContextTokens = getModelMaxTokens(this.modelOptions.model) ?? 4097;
+    this.maxContextTokens =
+      this.options.maxContextTokens ?? getModelMaxTokens(this.modelOptions.model) ?? 4097;
   }
   buildMessages() {}
   getTokenCount(str) {