@lobehub/chat 1.49.5 → 1.49.7

package/CHANGELOG.md

@@ -2,6 +2,56 @@
 
 # Changelog
 
+### [Version 1.49.7](https://github.com/lobehub/lobe-chat/compare/v1.49.6...v1.49.7)
+
+<sup>Released on **2025-02-01**</sup>
+
+#### 🐛 Bug Fixes
+
+- **misc**: Multiple deepseek-reasoner request errors.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### What's fixed
+
+- **misc**: Multiple deepseek-reasoner request errors, closes [#5601](https://github.com/lobehub/lobe-chat/issues/5601) ([71cc32b](https://github.com/lobehub/lobe-chat/commit/71cc32b))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
+### [Version 1.49.6](https://github.com/lobehub/lobe-chat/compare/v1.49.5...v1.49.6)
+
+<sup>Released on **2025-01-30**</sup>
+
+#### 🐛 Bug Fixes
+
+- **misc**: Support litellm reasoning streaming.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### What's fixed
+
+- **misc**: Support litellm reasoning streaming, closes [#5632](https://github.com/lobehub/lobe-chat/issues/5632) ([9942fb3](https://github.com/lobehub/lobe-chat/commit/9942fb3))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
 ### [Version 1.49.5](https://github.com/lobehub/lobe-chat/compare/v1.49.4...v1.49.5)
 
 <sup>Released on **2025-01-28**</sup>

package/changelog/v1.json

@@ -1,4 +1,22 @@
 [
+  {
+    "children": {
+      "fixes": [
+        "Multiple deepseek-reasoner request errors."
+      ]
+    },
+    "date": "2025-02-01",
+    "version": "1.49.7"
+  },
+  {
+    "children": {
+      "fixes": [
+        "Support litellm reasoning streaming."
+      ]
+    },
+    "date": "2025-01-30",
+    "version": "1.49.6"
+  },
   {
     "children": {
       "fixes": [

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lobehub/chat",
-  "version": "1.49.5",
+  "version": "1.49.7",
   "description": "Lobe Chat - an open-source, high-performance chatbot framework that supports speech synthesis, multimodal, and extensible Function Call plugin system. Supports one-click free deployment of your private ChatGPT/LLM web application.",
   "keywords": [
     "framework",

package/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);
+      });
+    });
   });
 });

package/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',

package/src/libs/agent-runtime/utils/streams/openai.test.ts

@@ -554,7 +554,7 @@ describe('OpenAIStream', () => {
   });
 
   describe('Reasoning', () => {
-    it('should handle reasoning event', async () => {
+    it('should handle reasoning event in official DeepSeek api', async () => {
       const data = [
         {
           id: '1',
@@ -722,6 +722,206 @@ describe('OpenAIStream', () => {
         chunks.push(decoder.decode(chunk, { stream: true }));
       }
 
+      expect(chunks).toEqual(
+        [
+          'id: 1',
+          'event: reasoning',
+          `data: ""\n`,
+          'id: 1',
+          'event: reasoning',
+          `data: "您好"\n`,
+          'id: 1',
+          'event: reasoning',
+          `data: "！"\n`,
+          'id: 1',
+          'event: text',
+          `data: "你好"\n`,
+          'id: 1',
+          'event: text',
+          `data: "很高兴"\n`,
+          'id: 1',
+          'event: text',
+          `data: "为您"\n`,
+          'id: 1',
+          'event: text',
+          `data: "提供"\n`,
+          'id: 1',
+          'event: text',
+          `data: "帮助。"\n`,
+          'id: 1',
+          'event: stop',
+          `data: "stop"\n`,
+        ].map((i) => `${i}\n`),
+      );
+    });
+    it('should handle reasoning in litellm', async () => {
+      const data = [
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { role: 'assistant', reasoning_content: '' },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { reasoning_content: '您好' },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { reasoning_content: '！' },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { content: '你好', reasoning_content: null },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { content: '很高兴', reasoning_cont: null },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { content: '为您', reasoning_content: null },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { content: '提供', reasoning_content: null },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { content: '帮助。', reasoning_content: null },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { content: '', reasoning_content: null },
+              logprobs: null,
+              finish_reason: 'stop',
+            },
+          ],
+          usage: {
+            prompt_tokens: 6,
+            completion_tokens: 104,
+            total_tokens: 110,
+            prompt_tokens_details: { cached_tokens: 0 },
+            completion_tokens_details: { reasoning_tokens: 70 },
+            prompt_cache_hit_tokens: 0,
+            prompt_cache_miss_tokens: 6,
+          },
+        },
+      ];
+
+      const mockOpenAIStream = new ReadableStream({
+        start(controller) {
+          data.forEach((chunk) => {
+            controller.enqueue(chunk);
+          });
+
+          controller.close();
+        },
+      });
+
+      const protocolStream = OpenAIStream(mockOpenAIStream);
+
+      const decoder = new TextDecoder();
+      const chunks = [];
+
+      // @ts-ignore
+      for await (const chunk of protocolStream) {
+        chunks.push(decoder.decode(chunk, { stream: true }));
+      }
+
       expect(chunks).toEqual(
         [
           'id: 1',

package/src/libs/agent-runtime/utils/streams/openai.ts

@@ -92,13 +92,18 @@ export const transformOpenAIStream = (
       return { data: item.delta.content, id: chunk.id, type: 'text' };
     }
 
+    // DeepSeek reasoner 会将 thinking 放在 reasoning_content 字段中
+    // litellm 处理 reasoning content 时 不会设定 content = null
+    if (
+      item.delta &&
+      'reasoning_content' in item.delta &&
+      typeof item.delta.reasoning_content === 'string'
+    ) {
+      return { data: item.delta.reasoning_content, id: chunk.id, type: 'reasoning' };
+    }
+
     // 无内容情况
     if (item.delta && item.delta.content === null) {
-      // deepseek reasoner 会将 thinking 放在 reasoning_content 字段中
-      if ('reasoning_content' in item.delta && typeof item.delta.reasoning_content === 'string') {
-        return { data: item.delta.reasoning_content, id: chunk.id, type: 'reasoning' };
-      }
-
       return { data: item.delta, id: chunk.id, type: 'data' };
     }