@lobehub/chat 1.26.2 → 1.26.3

package/CHANGELOG.md

@@ -2,6 +2,39 @@
 
 # Changelog
 
+### [Version 1.26.3](https://github.com/lobehub/lobe-chat/compare/v1.26.2...v1.26.3)
+
+<sup>Released on **2024-10-28**</sup>
+
+#### 🐛 Bug Fixes
+
+- **misc**: Fix Huggingface API interrupting when the output exceeds 140 tokens.
+
+#### 💄 Styles
+
+- **misc**: Remove SenseChat-Vision model, due to model limitation.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### What's fixed
+
+- **misc**: Fix Huggingface API interrupting when the output exceeds 140 tokens, closes [#4524](https://github.com/lobehub/lobe-chat/issues/4524) ([49acdd7](https://github.com/lobehub/lobe-chat/commit/49acdd7))
+
+#### Styles
+
+- **misc**: Remove SenseChat-Vision model, due to model limitation, closes [#4493](https://github.com/lobehub/lobe-chat/issues/4493) ([2aea033](https://github.com/lobehub/lobe-chat/commit/2aea033))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
 ### [Version 1.26.2](https://github.com/lobehub/lobe-chat/compare/v1.26.1...v1.26.2)
 
 <sup>Released on **2024-10-28**</sup>

package/docs/usage/agents/model.mdx

@@ -17,7 +17,7 @@ tags:
 ## ChatGPT
 
 - **gpt-3.5-turbo**: Currently the fastest generating ChatGPT model, it is faster but may sacrifice some text quality, with a context length of 4k.
-- **gpt-4**: ChatGPT 4.0 has improved language understanding and generation capabilities compared to 3.5. It can better understand context and context, and generate more accurate and natural responses. This is thanks to improvements in the GPT-4 model, including better language modeling and deeper semantic understanding, but it may be slower than other models, with a context length of 8k.
+- **gpt-4**: ChatGPT 4.0 has improved language understanding and generation capabilities compared to 3.5. It can better understand context and generate more accurate and natural responses. This is thanks to improvements in the GPT-4 model, including better language modeling and deeper semantic understanding, but it may be slower than other models, with a context length of 8k.
 - **gpt-4-32k**: Similar to gpt-4, the context limit is increased to 32k tokens, with a higher cost.
 
 ## Concept of Model Parameters

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lobehub/chat",
-  "version": "1.26.2",
+  "version": "1.26.3",
   "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/config/modelProviders/sensenova.ts

@@ -17,6 +17,8 @@ const SenseNova: ModelProviderCard = {
       },
       tokens: 131_072,
     },
+/*
+    // Not compatible with local mode (Not support Base64 Image)
     {
       description: '最新版本模型 (V5.5)，16K上下文长度，支持多图的输入，全面实现模型基础能力优化，在对象属性识别、空间关系、动作事件识别、场景理解、情感识别、逻辑常识推理和文本理解生成上都实现了较大提升。',
       displayName: 'SenseChat 5.5 Vision',
@@ -30,6 +32,7 @@ const SenseNova: ModelProviderCard = {
       tokens: 16_384,
       vision: true,
     },
+*/
     {
       description: '适用于快速问答、模型微调场景',
       displayName: 'SenseChat 5.0 Turbo',

package/src/libs/agent-runtime/huggingface/index.ts

@@ -21,8 +21,10 @@ export const LobeHuggingFaceAI = LobeOpenAICompatibleFactory({
   },
   customClient: {
     createChatCompletionStream: (client: HfInference, payload, instance) => {
+      const { max_tokens = 4096} = payload;
       const hfRes = client.chatCompletionStream({
         endpointUrl: instance.baseURL,
+        max_tokens: max_tokens,
         messages: payload.messages,
         model: payload.model,
         stream: true,