feat: support gemini-pro-vision

app/client/platforms/google.ts

@@ -13,6 +13,13 @@ import {
   LLMUsage,
 } from "../api";
 import { useAccessStore, useAppConfig, useChatStore } from "@/app/store";
+import axios from "axios";
+
+const getImageBase64Data = async (url: string) => {
+  const response = await axios.get(url, { responseType: "arraybuffer" });
+  const base64 = Buffer.from(response.data, "binary").toString("base64");
+  return base64;
+};
 
 export class GeminiProApi implements LLMApi {
   toolAgentChat(options: AgentChatOptions): Promise<void> {
@@ -28,11 +35,32 @@ export class GeminiProApi implements LLMApi {
     );
   }
   async chat(options: ChatOptions): Promise<void> {
-    const apiClient = this;
+    const messages: any[] = [];
-    const messages = options.messages.map((v) => ({
+    if (options.config.model.includes("vision")) {
-      role: v.role.replace("assistant", "model").replace("system", "user"),
+      for (const v of options.messages) {
-      parts: [{ text: v.content }],
+        let message: any = {
-    }));
+          role: v.role.replace("assistant", "model").replace("system", "user"),
+          parts: [{ text: v.content }],
+        };
+        if (v.image_url) {
+          var base64Data = await getImageBase64Data(v.image_url);
+          message.parts.push({
+            inline_data: {
+              mime_type: "image/jpeg",
+              data: base64Data,
+            },
+          });
+        }
+        messages.push(message);
+      }
+    } else {
+      options.messages.map((v) =>
+        messages.push({
+          role: v.role.replace("assistant", "model").replace("system", "user"),
+          parts: [{ text: v.content }],
+        }),
+      );
+    }
 
     // google requires that role in neighboring messages must not be the same
     for (let i = 0; i < messages.length - 1; ) {
@@ -92,7 +120,9 @@ export class GeminiProApi implements LLMApi {
     const controller = new AbortController();
     options.onController?.(controller);
     try {
-      const chatPath = this.path(Google.ChatPath);
+      const chatPath = this.path(
+        Google.ChatPath.replace("{{model}}", options.config.model),
+      );
       const chatPayload = {
         method: "POST",
         body: JSON.stringify(requestPayload),

app/client/platforms/openai.ts

@@ -140,10 +140,9 @@ export class ChatGPTApi implements LLMApi {
       presence_penalty: modelConfig.presence_penalty,
       frequency_penalty: modelConfig.frequency_penalty,
       top_p: modelConfig.top_p,
-      max_tokens:
+      max_tokens: modelConfig.model.includes("vision")
-        modelConfig.model == "gpt-4-vision-preview"
+        ? modelConfig.max_tokens
-          ? modelConfig.max_tokens
+        : null,
-          : null,
       // max_tokens: Math.max(modelConfig.max_tokens, 1024),
       // Please do not ask me why not send max_tokens, no reason, this param is just shit, I dont want to explain anymore.
     };

app/components/chat.tsx

@@ -538,7 +538,7 @@ export function ChatActions(props: {
         }
       }
     };
-    if (currentModel === "gpt-4-vision-preview") {
+    if (currentModel.includes("vision")) {
       window.addEventListener("paste", onPaste);
       return () => {
         window.removeEventListener("paste", onPaste);
@@ -620,7 +620,7 @@ export function ChatActions(props: {
               icon={usePlugins ? <EnablePluginIcon /> : <DisablePluginIcon />}
             />
           )}
-        {currentModel == "gpt-4-vision-preview" && (
+        {currentModel.includes("vision") && (
           <ChatAction
             onClick={selectImage}
             text="选择图片"
@@ -1412,7 +1412,7 @@ function _Chat() {
                       defaultShow={i >= messages.length - 6}
                     />
                   </div>
-                  {!isUser && message.model == "gpt-4-vision-preview" && (
+                  {!isUser && message.model?.includes("vision") && (
                     <div
                       className={[
                         styles["chat-message-actions"],

app/constant.ts

@@ -91,7 +91,7 @@ export const Azure = {
 
 export const Google = {
   ExampleEndpoint: "https://generativelanguage.googleapis.com/",
-  ChatPath: "v1beta/models/gemini-pro:generateContent",
+  ChatPath: "v1beta/models/{{model}}:generateContent",
 };
 
 export const DEFAULT_INPUT_TEMPLATE = `{{input}}`; // input / time / model / lang
@@ -253,6 +253,15 @@ export const DEFAULT_MODELS = [
       providerType: "google",
     },
   },
+  {
+    name: "gemini-pro-vision",
+    available: true,
+    provider: {
+      id: "google",
+      providerName: "Google",
+      providerType: "google",
+    },
+  },
 ] as const;
 
 export const CHAT_PAGE_SIZE = 15;