Merge pull request #5173 from ConnectAI-E/feature/dalle

add dalle3 model

app/client/api.ts

@@ -6,7 +6,7 @@ import {
   ServiceProvider,
 } from "../constant";
 import { ChatMessage, ModelType, useAccessStore, useChatStore } from "../store";
-import { ChatGPTApi } from "./platforms/openai";
+import { ChatGPTApi, DalleRequestPayload } from "./platforms/openai";
 import { GeminiProApi } from "./platforms/google";
 import { ClaudeApi } from "./platforms/anthropic";
 import { ErnieApi } from "./platforms/baidu";
@@ -42,6 +42,7 @@ export interface LLMConfig {
   stream?: boolean;
   presence_penalty?: number;
   frequency_penalty?: number;
+  size?: DalleRequestPayload["size"];
 }
 
 export interface ChatOptions {

app/client/platforms/openai.ts

@@ -11,8 +11,13 @@ import {
 } from "@/app/constant";
 import { useAccessStore, useAppConfig, useChatStore } from "@/app/store";
 import { collectModelsWithDefaultModel } from "@/app/utils/model";
-import { preProcessImageContent } from "@/app/utils/chat";
+import {
+  preProcessImageContent,
+  uploadImage,
+  base64Image2Blob,
+} from "@/app/utils/chat";
 import { cloudflareAIGatewayUrl } from "@/app/utils/cloudflare";
+import { DalleSize } from "@/app/typing";
 
 import {
   ChatOptions,
@@ -33,6 +38,7 @@ import {
   getMessageTextContent,
   getMessageImages,
   isVisionModel,
+  isDalle3 as _isDalle3,
 } from "@/app/utils";
 
 export interface OpenAIListModelResponse {
@@ -58,6 +64,14 @@ export interface RequestPayload {
   max_tokens?: number;
 }
 
+export interface DalleRequestPayload {
+  model: string;
+  prompt: string;
+  response_format: "url" | "b64_json";
+  n: number;
+  size: DalleSize;
+}
+
 export class ChatGPTApi implements LLMApi {
   private disableListModels = true;
 
@@ -100,20 +114,31 @@ export class ChatGPTApi implements LLMApi {
     return cloudflareAIGatewayUrl([baseUrl, path].join("/"));
   }
 
-  extractMessage(res: any) {
-    return res.choices?.at(0)?.message?.content ?? "";
+  async extractMessage(res: any) {
+    if (res.error) {
+      return "```\n" + JSON.stringify(res, null, 4) + "\n```";
+    }
+    // dalle3 model return url, using url create image message
+    if (res.data) {
+      let url = res.data?.at(0)?.url ?? "";
+      const b64_json = res.data?.at(0)?.b64_json ?? "";
+      if (!url && b64_json) {
+        // uploadImage
+        url = await uploadImage(base64Image2Blob(b64_json, "image/png"));
+      }
+      return [
+        {
+          type: "image_url",
+          image_url: {
+            url,
+          },
+        },
+      ];
+    }
+    return res.choices?.at(0)?.message?.content ?? res;
   }
 
   async chat(options: ChatOptions) {
-    const visionModel = isVisionModel(options.config.model);
-    const messages: ChatOptions["messages"] = [];
-    for (const v of options.messages) {
-      const content = visionModel
-        ? await preProcessImageContent(v.content)
-        : getMessageTextContent(v);
-      messages.push({ role: v.role, content });
-    }
-
     const modelConfig = {
       ...useAppConfig.getState().modelConfig,
       ...useChatStore.getState().currentSession().mask.modelConfig,
@@ -123,26 +148,52 @@ export class ChatGPTApi implements LLMApi {
       },
     };
 
-    const requestPayload: RequestPayload = {
-      messages,
-      stream: options.config.stream,
-      model: modelConfig.model,
-      temperature: modelConfig.temperature,
-      presence_penalty: modelConfig.presence_penalty,
-      frequency_penalty: modelConfig.frequency_penalty,
-      top_p: modelConfig.top_p,
-      // 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.
-    };
+    let requestPayload: RequestPayload | DalleRequestPayload;
 
-    // add max_tokens to vision model
-    if (visionModel && modelConfig.model.includes("preview")) {
-      requestPayload["max_tokens"] = Math.max(modelConfig.max_tokens, 4000);
+    const isDalle3 = _isDalle3(options.config.model);
+    if (isDalle3) {
+      const prompt = getMessageTextContent(
+        options.messages.slice(-1)?.pop() as any,
+      );
+      requestPayload = {
+        model: options.config.model,
+        prompt,
+        // URLs are only valid for 60 minutes after the image has been generated.
+        response_format: "b64_json", // using b64_json, and save image in CacheStorage
+        n: 1,
+        size: options.config?.size ?? "1024x1024",
+      };
+    } else {
+      const visionModel = isVisionModel(options.config.model);
+      const messages: ChatOptions["messages"] = [];
+      for (const v of options.messages) {
+        const content = visionModel
+          ? await preProcessImageContent(v.content)
+          : getMessageTextContent(v);
+        messages.push({ role: v.role, content });
+      }
+
+      requestPayload = {
+        messages,
+        stream: options.config.stream,
+        model: modelConfig.model,
+        temperature: modelConfig.temperature,
+        presence_penalty: modelConfig.presence_penalty,
+        frequency_penalty: modelConfig.frequency_penalty,
+        top_p: modelConfig.top_p,
+        // 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.
+      };
+
+      // add max_tokens to vision model
+      if (visionModel && modelConfig.model.includes("preview")) {
+        requestPayload["max_tokens"] = Math.max(modelConfig.max_tokens, 4000);
+      }
     }
 
     console.log("[Request] openai payload: ", requestPayload);
 
-    const shouldStream = !!options.config.stream;
+    const shouldStream = !isDalle3 && !!options.config.stream;
     const controller = new AbortController();
     options.onController?.(controller);
 
@@ -168,13 +219,15 @@ export class ChatGPTApi implements LLMApi {
             model?.provider?.providerName === ServiceProvider.Azure,
         );
         chatPath = this.path(
-          Azure.ChatPath(
+          (isDalle3 ? Azure.ImagePath : Azure.ChatPath)(
             (model?.displayName ?? model?.name) as string,
             useCustomConfig ? useAccessStore.getState().azureApiVersion : "",
           ),
         );
       } else {
-        chatPath = this.path(OpenaiPath.ChatPath);
+        chatPath = this.path(
+          isDalle3 ? OpenaiPath.ImagePath : OpenaiPath.ChatPath,
+        );
       }
       const chatPayload = {
         method: "POST",
@@ -186,7 +239,7 @@ export class ChatGPTApi implements LLMApi {
       // make a fetch request
       const requestTimeoutId = setTimeout(
         () => controller.abort(),
-        REQUEST_TIMEOUT_MS,
+        isDalle3 ? REQUEST_TIMEOUT_MS * 2 : REQUEST_TIMEOUT_MS, // dalle3 using b64_json is slow.
       );
 
       if (shouldStream) {
@@ -317,7 +370,7 @@ export class ChatGPTApi implements LLMApi {
         clearTimeout(requestTimeoutId);
 
         const resJson = await res.json();
-        const message = this.extractMessage(resJson);
+        const message = await this.extractMessage(resJson);
         options.onFinish(message);
       }
     } catch (e) {