Merge ecc65fa775 into c30ddfbb07

Fix: Encourage markdown inclusion in model responses for o1/o3

app/api/common.ts

@@ -1,6 +1,6 @@
 import { NextRequest, NextResponse } from "next/server";
 import { getServerSideConfig } from "../config/server";
-import { OPENAI_BASE_URL, ServiceProvider } from "../constant";
+import { OPENAI_BASE_URL, ServiceProvider, OpenaiPath } from "../constant";
 import { cloudflareAIGatewayUrl } from "../utils/cloudflare";
 import { getModelProvider, isModelNotavailableInServer } from "../utils/model";
 
@@ -26,8 +26,8 @@ export async function requestOpenai(req: NextRequest) {
     authValue = req.headers.get("Authorization") ?? "";
     authHeaderName = "Authorization";
   }
-
   let path = `${req.nextUrl.pathname}`.replaceAll("/api/openai/", "");
+  let isChatRequest = path.includes(OpenaiPath.ChatPath);
 
   let baseUrl =
     (isAzure ? serverConfig.azureUrl : serverConfig.baseUrl) || OPENAI_BASE_URL;
@@ -117,14 +117,14 @@ export async function requestOpenai(req: NextRequest) {
       const jsonBody = JSON.parse(clonedBody) as { model?: string };
 
       // not undefined and is false
-      if (
+      if ( isChatRequest &&
         isModelNotavailableInServer(
           serverConfig.customModels,
           jsonBody?.model as string,
           [
             ServiceProvider.OpenAI,
             ServiceProvider.Azure,
-            jsonBody?.model as string, // support provider-unspecified model
+            "custom" as string, // support provider-unspecified model
           ],
         )
       ) {

app/client/platforms/openai.ts

@@ -56,7 +56,7 @@ export interface OpenAIListModelResponse {
 
 export interface RequestPayload {
   messages: {
-    role: "system" | "user" | "assistant";
+    role: "developer" | "system" | "user" | "assistant";
     content: string | MultimodalContent[];
   }[];
   stream?: boolean;
@@ -238,8 +238,16 @@ export class ChatGPTApi implements LLMApi {
         // Please do not ask me why not send max_tokens, no reason, this param is just shit, I dont want to explain anymore.
       };
 
-      // O1 使用 max_completion_tokens 控制token数 (https://platform.openai.com/docs/guides/reasoning#controlling-costs)
       if (isO1OrO3) {
+        // by default the o1/o3 models will not attempt to produce output that includes markdown formatting
+        // manually add "Formatting re-enabled" developer message to encourage markdown inclusion in model responses
+        // (https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/reasoning?tabs=python-secure#markdown-output)
+        requestPayload["messages"].unshift({
+          role: "developer",
+          content: "Formatting re-enabled",
+        });
+
+        // o1/o3 uses max_completion_tokens to control the number of tokens (https://platform.openai.com/docs/guides/reasoning#controlling-costs)
         requestPayload["max_completion_tokens"] = modelConfig.max_tokens;
       }
 

app/constant.ts

@@ -633,6 +633,18 @@ const xAIModes = [
   "grok-2-vision-1212",
   "grok-2-vision",
   "grok-2-vision-latest",
+  "grok-3-mini-fast-beta",
+  "grok-3-mini-fast",
+  "grok-3-mini-fast-latest",
+  "grok-3-mini-beta",
+  "grok-3-mini",
+  "grok-3-mini-latest",
+  "grok-3-fast-beta",
+  "grok-3-fast",
+  "grok-3-fast-latest",
+  "grok-3-beta",
+  "grok-3",
+  "grok-3-latest",
 ];
 
 const chatglmModels = [

app/utils/model.ts

@@ -66,8 +66,8 @@ export function collectModelTable(
 
   // default models
   models.forEach((m) => {
-    // using <modelName>@<providerId> as fullName
-    modelTable[`${m.name}@${m?.provider?.id}`] = {
+    // using <modelName>@<providerType> as fullName
+    modelTable[`${m.name}@${m?.provider?.providerType}`] = {
       ...m,
       displayName: m.name, // 'provider' is copied over if it exists
     };
@@ -121,12 +121,14 @@ export function collectModelTable(
           if (displayName && provider.providerName == "ByteDance") {
             [customModelName, displayName] = [displayName, customModelName];
           }
-          modelTable[`${customModelName}@${provider?.id}`] = {
+          modelTable[`${customModelName}@${provider?.providerType}`] = {
             name: customModelName,
             displayName: displayName || customModelName,
             available,
             provider, // Use optional chaining
-            sorted: CustomSeq.next(`${customModelName}@${provider?.id}`),
+            sorted: CustomSeq.next(
+              `${customModelName}@${provider?.providerType}`,
+            ),
           };
         }
       }

test/model-available.test.ts

@@ -53,7 +53,7 @@ describe("isModelNotavailableInServer", () => {
     expect(result).toBe(true);
   });
 
-  // FIXME: 这个测试用例有问题，需要修复
+  // FIXME: 这个测试用例有问题，需要修复 ？？？
   //   test("support passing multiple providers, model available on one of the providers will return false", () => {
   //     const customModels = "-all,gpt-4@google";
   //     const modelName = "gpt-4";
@@ -69,7 +69,19 @@ describe("isModelNotavailableInServer", () => {
   test("test custom model without setting provider", () => {
     const customModels = "-all,mistral-large";
     const modelName = "mistral-large";
-    const providerNames = modelName;
+    const providerNames = "custom";
+    const result = isModelNotavailableInServer(
+      customModels,
+      modelName,
+      providerNames,
+    );
+    expect(result).toBe(false);
+  });
+
+  test("test custom model with non-standard provider", () => {
+    const customModels = "-all,deepseek-chat@DeepSeek";
+    const modelName = "deepseek-chat";
+    const providerNames = "custom";
     const result = isModelNotavailableInServer(
       customModels,
       modelName,