🐛 fix: Multiple deepseek-reasoner request errors (#5601)

* Update index.ts

* Update index.ts

* Update index.test.ts

Author: sxjeru

src/libs/agent-runtime/deepseek/index.test.ts

@@ -4,12 +4,15 @@ import { Mock, afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
 
 import {
   ChatStreamCallbacks,
+  ChatStreamPayload,
+  LLMRoleType,
   LobeOpenAICompatibleRuntime,
   ModelProvider,
 } from '@/libs/agent-runtime';
 
 import * as debugStreamModule from '../utils/debugStream';
 import { LobeDeepSeekAI } from './index';
+import { LobeOpenAICompatibleFactory } from '../utils/openaiCompatibleFactory';
 
 const provider = ModelProvider.DeepSeek;
 const defaultBaseURL = 'https://api.deepseek.com/v1';
@@ -22,6 +25,17 @@ vi.spyOn(console, 'error').mockImplementation(() => {});
 
 let instance: LobeOpenAICompatibleRuntime;
 
+const createDeepSeekAIInstance = () => new LobeDeepSeekAI({ apiKey: 'test' });
+
+const mockSuccessfulChatCompletion = () => {
+  vi.spyOn(instance['client'].chat.completions, 'create').mockResolvedValue({
+    id: 'cmpl-mock',
+    object: 'chat.completion',
+    created: Date.now(),
+    choices: [{ index: 0, message: { role: 'assistant', content: 'Mock response' }, finish_reason: 'stop' }],
+  } as any);
+};
+
 beforeEach(() => {
   instance = new LobeDeepSeekAI({ apiKey: 'test' });
 
@@ -251,5 +265,126 @@ describe('LobeDeepSeekAI', () => {
         process.env.DEBUG_DEEPSEEK_CHAT_COMPLETION = originalDebugValue;
       });
     });
+
+    describe('deepseek-reasoner', () => {
+      beforeEach(() => {
+        instance = createDeepSeekAIInstance();
+        mockSuccessfulChatCompletion();
+      });
+
+      it('should insert a user message if the first message is from assistant', async () => {
+        const payloadMessages = [{ content: 'Hello', role: 'assistant' as LLMRoleType }];
+        const expectedMessages = [
+          { content: '', role: 'user' },
+          ...payloadMessages,
+        ];
+
+        const payload: ChatStreamPayload = {
+          messages: payloadMessages,
+          model: 'deepseek-reasoner',
+          temperature: 0,
+        };
+
+        await instance.chat(payload);
+
+        expect(instance['client'].chat.completions.create).toHaveBeenCalled();
+        const actualArgs = (instance['client'].chat.completions.create as Mock).mock.calls[0];
+        const actualMessages = actualArgs[0].messages;
+        expect(actualMessages).toEqual(expectedMessages);
+      });
+
+      it('should insert a user message if the first message is from assistant (with system summary)', async () => {
+        const payloadMessages = [
+          { content: 'System summary', role: 'system' as LLMRoleType },
+          { content: 'Hello', role: 'assistant' as LLMRoleType },
+        ];
+        const expectedMessages = [
+          { content: 'System summary', role: 'system' },
+          { content: '', role: 'user' },
+          { content: 'Hello', role: 'assistant' },
+        ];
+
+        const payload: ChatStreamPayload = {
+          messages: payloadMessages,
+          model: 'deepseek-reasoner',
+          temperature: 0,
+        };
+
+        await instance.chat(payload);
+
+        expect(instance['client'].chat.completions.create).toHaveBeenCalled();
+        const actualArgs = (instance['client'].chat.completions.create as Mock).mock.calls[0];
+        const actualMessages = actualArgs[0].messages;
+        expect(actualMessages).toEqual(expectedMessages);
+      });
+
+      it('should insert alternating roles if messages do not alternate', async () => {
+        const payloadMessages = [
+          { content: 'user1', role: 'user' as LLMRoleType },
+          { content: 'user2', role: 'user' as LLMRoleType },
+          { content: 'assistant1', role: 'assistant' as LLMRoleType },
+          { content: 'assistant2', role: 'assistant' as LLMRoleType },
+        ];
+        const expectedMessages = [
+          { content: 'user1', role: 'user' },
+          { content: '', role: 'assistant' },
+          { content: 'user2', role: 'user' },
+          { content: 'assistant1', role: 'assistant' },
+          { content: '', role: 'user' },
+          { content: 'assistant2', role: 'assistant' },
+        ];
+
+        const payload: ChatStreamPayload = {
+          messages: payloadMessages,
+          model: 'deepseek-reasoner',
+          temperature: 0,
+        };
+
+        await instance.chat(payload);
+
+        expect(instance['client'].chat.completions.create).toHaveBeenCalled();
+        const actualArgs = (instance['client'].chat.completions.create as Mock).mock.calls[0];
+        const actualMessages = actualArgs[0].messages;
+        expect(actualMessages).toEqual(expectedMessages);
+      });
+
+      it('complex condition', async () => {
+        const payloadMessages = [
+          { content: 'system', role: 'system' as LLMRoleType },
+          { content: 'assistant', role: 'assistant' as LLMRoleType },
+          { content: 'user1', role: 'user' as LLMRoleType },
+          { content: 'user2', role: 'user' as LLMRoleType },
+          { content: 'user3', role: 'user' as LLMRoleType },
+          { content: 'assistant1', role: 'assistant' as LLMRoleType },
+          { content: 'assistant2', role: 'assistant' as LLMRoleType },
+        ];
+        const expectedMessages = [
+          { content: 'system', role: 'system' },
+          { content: '', role: 'user' },
+          { content: 'assistant', role: 'assistant' },
+          { content: 'user1', role: 'user' },
+          { content: '', role: 'assistant' },
+          { content: 'user2', role: 'user' },
+          { content: '', role: 'assistant' },
+          { content: 'user3', role: 'user' },
+          { content: 'assistant1', role: 'assistant' },
+          { content: '', role: 'user' },
+          { content: 'assistant2', role: 'assistant' },
+        ];
+
+        const payload: ChatStreamPayload = {
+          messages: payloadMessages,
+          model: 'deepseek-reasoner',
+          temperature: 0,
+        };
+
+        await instance.chat(payload);
+
+        expect(instance['client'].chat.completions.create).toHaveBeenCalled();
+        const actualArgs = (instance['client'].chat.completions.create as Mock).mock.calls[0];
+        const actualMessages = actualArgs[0].messages;
+        expect(actualMessages).toEqual(expectedMessages);
+      });
+    });
   });
 });

src/libs/agent-runtime/deepseek/index.ts

@@ -12,24 +12,49 @@ export interface DeepSeekModelCard {
 export const LobeDeepSeekAI = LobeOpenAICompatibleFactory({
   baseURL: 'https://api.deepseek.com/v1',
   chatCompletion: {
-    handlePayload: ({ frequency_penalty, model, presence_penalty, temperature, top_p, ...payload }: ChatStreamPayload) =>
-      ({
+    handlePayload: ({ frequency_penalty, messages, model, presence_penalty, temperature, top_p, ...payload }: ChatStreamPayload) => {
+      // github.com/lobehub/lobe-chat/pull/5548
+      let filteredMessages = messages.filter(message => message.role !== 'system');
+
+      if (filteredMessages.length > 0 && filteredMessages[0].role === 'assistant') {
+        filteredMessages.unshift({ content: "", role: "user" });
+      }
+
+      let lastRole = '';
+      for (let i = 0; i < filteredMessages.length; i++) {
+        const message = filteredMessages[i];
+        if (message.role === lastRole) {
+          const newRole = lastRole === 'assistant' ? 'user' : 'assistant';
+          filteredMessages.splice(i, 0, { content: "", role: newRole });
+          i++;
+        }
+        lastRole = message.role;
+      }
+
+      if (messages.length > 0 && messages[0].role === 'system') {
+        filteredMessages.unshift(messages[0]);
+      }
+
+      return {
         ...payload,
         model,
         ...(model === 'deepseek-reasoner'
           ? {
               frequency_penalty: undefined,
+              messages: filteredMessages,
               presence_penalty: undefined,
               temperature: undefined,
               top_p: undefined,
             }
           : {
               frequency_penalty,
+              messages,
               presence_penalty,
               temperature,
               top_p,
             }),
-      }) as OpenAI.ChatCompletionCreateParamsStreaming,
+      } as OpenAI.ChatCompletionCreateParamsStreaming;
+    },
   },
   debug: {
     chatCompletion: () => process.env.DEBUG_DEEPSEEK_CHAT_COMPLETION === '1',