Skip to content

ChatSpeed v1.1.23

Choose a tag to compare

View all tags
@github-actions github-actions released this 23 Jan 02:03
v1.1.23
e840358

[1.1.23] - 2026-01-22

🚀 New Features

  • Model-Specific Custom Body Parameters: Added support for configuring custom JSON parameters (e.g., enable_thinking) at the specific model ID level.
  • Dynamic Group Switching (/switch): Introduced a unified /switch route prefix that dynamically resolves requests based on the currently "Active" proxy group.
  • System Prompt Keyword Replacement: Implemented KV-based replacement rules in proxy groups to automatically correct System Prompts before forwarding.
  • Proxy Group Batch Update: Supports synchronizing prompt injection, text replacement, and tool filtering across multiple groups in one transaction.

🪄 Improvements

  • Refined Parameter Injection Hierarchy: Established a strict "Model Patch > User UI Intent > Engine Default" priority system.
  • Smart Type Conversion: Automatically converts string values like "true"/"false" to Booleans and numeric strings to Numbers for provider compatibility.
  • UI Refactoring: Upgraded the Proxy Group dialog to a tabbed layout, matching the AI Engine settings style and fixing button visibility issues.
  • Enhanced Robustness: process_custom_params now intelligently handles various nested data structures and edge cases.

🐞 Bug Fixes

  • ModelScope/Qwen Compatibility: Automatically forces enable_thinking to false for non-streaming requests to prevent 400 Bad Request errors.
  • UI State Persistence: Fixed an issue where internal chat might lose real-time parameters (e.g., temperature, stop sequences) during protocol conversion.
  • Element Plus Adaption: Resolved deprecation warnings for el-radio component APIs.
  • Security Hardening: Removed HTML tags from i18n messages to eliminate XSS warnings.

[1.1.23] - 2026-01-22

🚀 新特性

  • 模型级自定义 Body 参数：支持在具体的模型 ID 级别配置附加参数（如魔塔的 enable_thinking）。
  • 动态分组切换 (/switch)：新增统一的 /switch 路由前缀，代理会根据当前“激活”的分组动态转发请求，无需在 IDE 中频繁修改路径。
  • 提示词关键词替换：在代理分组中支持 KV 形式的文本替换规则，在请求发出前自动修正系统提示词（System Prompt）。
  • 代理分组批量更新：支持一键同步多个分组的提示词注入、文本替换及工具过滤设置。

🪄 改进

  • 参数注入优先级金字塔：确立了“模型补丁 > 用户 UI 设置 > 引擎默认值”的严密注入逻辑。
  • 智能类型转换：自定义参数支持自动将字符串 "true"/"false" 转换为布尔值，数字字符串转换为数值，确保供应商兼容性。
  • UI 重构：代理分组编辑界面升级为选项卡（Tab）布局，风格与“AI 引擎”设置统一，解决了长列表遮挡按钮的问题。
  • 增强鲁棒性：process_custom_params 现在可以自动识别并处理多种嵌套级别的数据结构。

🐞 修复

  • 魔塔/Qwen 兼容性补丁：自动在非流式请求时强制关闭 enable_thinking，规避 400 报错。
  • UI 状态同步：修复了内部聊天模块在转换过程中可能丢失 UI 实时参数（如温度、停止符）的问题。
  • Element Plus 适配：修正了 el-radio 等组件的过时 API 用法警告。
  • 安全增强：移除了国际化消息中的 HTML 标签，解决 XSS 安全警告。
Assets 17
Loading