Skip to content

0.2.1

Choose a tag to compare

View all tags
@Dreaminko Dreaminko released this 09 Oct 11:19
· 53 commits to main since this release
0.2.1
4959216

This release makes local Qwen runtime downloads optional, improves local model management and settings editing, and fixes translation model selection and Alibaba Cloud thinking compatibility.

New Features

  1. Qwen runtime downloads: The Qwen runtime is now downloaded on demand through recognition settings, with progress, verification, cancellation, and retry controls, instead of being included in the installer; cloud recognition does not require it.
  2. Local OCR model cleanup: Added deletion of downloaded PP-OCRv6 models from OCR settings, with deletion blocked while a download or local recognition is in progress.

Fixes & Improvements

  1. Model deletion: Added in-app confirmation dialogs for Whisper, Qwen, and OCR model deletion, and allowed deletion of the selected local recognition model with automatic switching to another installed model or stopping recognition when none is available.
  2. Settings editing: Fixed focus loss while editing speech and OCR translation model names, language preset names, and translation context limits, with changes saved when leaving the field and invalid context limits restored to their previous values.
  3. Translation model choices: Each translation target now remembers the model selected for each API profile and restores it when switching back, including manually entered model names.
  4. Alibaba Cloud thinking: Added thinking controls for DeepSeek V4.1 and Vanchin DeepSeek V4/V4.1 models, reserved more output tokens when thinking is enabled for Alibaba Cloud and Qwen translation, and now report responses cut off by the output token limit as errors.

Development Notes

  1. Optional GPU development tools: Default development builds and Rust checks now support cloud features and local CPU recognition without requiring the Vulkan SDK or Ninja; use npm run dev:vulkan or scripts/check-rust.ps1 -Vulkan for Vulkan development, while release packages retain Vulkan support.
  2. Windows build paths: Shortened CI and release build output paths to avoid MSVC path-length failures when compiling Vulkan shaders.

本次更新将本地 Qwen 运行组件改为按需下载，改进本地模型管理与设置编辑，并修复翻译模型选择和阿里云思考模式兼容性问题。

新功能

  1. Qwen 运行组件下载： Qwen 运行组件改为通过识别设置按需下载，支持进度显示、校验、取消和重试，不再随安装包分发，云端识别无需下载这些组件。
  2. 本地 OCR 模型清理： 新增在 OCR 设置中删除已下载的 PP-OCRv6 模型，下载或本地识别进行期间会阻止删除。

修复与改进

  1. 模型删除： 为 Whisper、Qwen 和 OCR 模型删除加入应用内确认弹窗，并支持删除当前选中的本地识别模型，自动切换到其他已安装模型，没有可用模型时停止识别。
  2. 设置编辑： 修复编辑语音与 OCR 翻译模型名称、语言预设名称和翻译上下文限制时输入框失去焦点的问题，离开输入框后再保存，并将无效的上下文限制恢复为原值。
  3. 翻译模型选择： 每个翻译目标会分别记住各 API 配置所选的模型，切换回来时恢复原选择，也支持手动输入的模型名称。
  4. 阿里云思考模式： 为 DeepSeek V4.1 和 Vanchin DeepSeek V4/V4.1 模型加入思考模式控制，为启用思考的阿里云与 Qwen 翻译预留更多输出 token，并将因输出 token 达到上限而截断的响应报告为错误。

开发说明

  1. 可选 GPU 开发工具： 默认开发构建与 Rust 检查支持云端功能和本地 CPU 识别，无需 Vulkan SDK 或 Ninja；开发 Vulkan 功能时可使用 npm run dev:vulkan 或 scripts/check-rust.ps1 -Vulkan，发布安装包仍保留 Vulkan 支持。
  2. Windows 构建路径： 缩短 CI 与发布构建的输出路径，避免编译 Vulkan 着色器时触发 MSVC 路径长度限制。

Full Changelog / 完整变更： 0.2.0...0.2.1

Assets 9
Loading