Skip to content

1.1.0

Latest
Latest

Choose a tag to compare

View all tags
@ArkaneFans ArkaneFans released this 17 Aug 13:26
· 3 commits to main since this release
1.1.0
32f227b

v1.1.0

新增

  • MNN 推理引擎：新增 MNN (https://github.com/alibaba/MNN) 引擎支持，与 llama.cpp 并列可选。可直接下载、导入、重命名 MNN 模型，原生后端由 mnn_engine (https://pub.dev/packages/mnn_engine) 插件提供。
  • 应用内模型下载：支持在应用内浏览并下载 MNN 模型，GGUF / MNN 两种格式的发现、下载、管理体验统一。
  • 聊天页引擎快切：聊天页服务器按钮新增引擎选择弹窗，无需跳转即可切换 llama.cpp / MNN 和对应模型。
  • MNN 引擎测试页：内置 MNN 测试页面，支持工具调用（Tool Calling）与多模态测试。

改进

  • llama.cpp 引擎改为非路由模式，两种引擎行为一致：均在启动前选定模型，切换模型由应用自动编排。
  • 升级 llama-server 运行库，并新增官方 GitHub Actions 构建工作流，便于后续跟进 llama.cpp 上游更新。
  • 优化服务器页与模型搜索页 UI，精简多余文案，更新筛选条件。
  • 优化 MNN 启动路径，切换引擎时不再出现端口占用。

修复

  • 修复模型下载进度显示与 MNN 模型导入失败的问题。
  • 修复切换引擎时端口被占用导致服务启动失败的问题。
  • 修复日志页长文本不能自动换行、模型搜索页 UI 对齐与搜索异常等问题。
  • 模型加载超时设为 60 秒，超时会在聊天客户端中显示明确错误。

English

Added

  • MNN inference engine: ServLlama now supports MNN (https://github.com/alibaba/MNN) alongside llama.cpp. Download, import, and rename MNN models directly in the app; the native backend is provided by the mnn_engine (https://pub.dev/packages/mnn_engine) plugin.
  • In-app model downloads: browse and download MNN models in the app, with a unified discovery and download experience for both GGUF and MNN formats.
  • Quick engine switching in chat: the server button on the Chat page now opens an engine picker, so you can switch between llama.cpp / MNN and their models without leaving the page.
  • MNN engine test page: a built-in test page for MNN covering tool calling and multimodal input.

Changed

  • The llama.cpp engine now runs in non-routing mode, aligning its behavior with MNN: a model is selected before startup, and model switching is orchestrated by the app.
  • Updated the bundled llama-server libraries and added an official GitHub Actions workflow to track future llama.cpp updates.
  • Refined the Server and model search pages: cleaner UI, less clutter, and updated filter options.
  • Optimized the MNN startup path; switching engines no longer causes port conflicts.

Fixed

  • Fixed model download progress display and MNN model import failures.
  • Fixed port occupation that could prevent the server from starting when switching engines.
  • Fixed missing automatic line wrapping on the log page, UI misalignment and search issues on the model search page.
  • Model loading now times out after 60 seconds with a clear error shown in the chat client.

What's Changed

Full Changelog: 1.0.10...1.1.0

Contributors

ArkaneFans
Assets 3
Loading