Skip to content

GestureSign V2 18.3.2

Choose a tag to compare

@Tomclanc Tomclanc released this 29 Sep 12:18
7daf27d

18.3.2

  • AI 默认提供轻量 CPU 版,保留本地采样、训练、评分、纠错和 AI 否决。x64 下载约 0.78 MiB,解压约 3.15 MiB;原版分别为 59.65 MiB 和 147.17 MiB。
  • 可选 NPU / GPU / CPU 硬件加速版,x64 下载约 24.49 MiB,解压约 70.55 MiB。NPU 厂商提供程序、驱动和缓存不包含在此数字内;当前采用 NPU → GPU → CPU 回退,不会同时调用 NPU 和 GPU 联合推理。
  • 两种 AI 组件复用对应架构的系统 .NET 10 Runtime,减少重复运行库。硬件版已通过 NVIDIA GPU 评分校验。
  • 保留已安装 18.3.1 AI 组件的自动恢复兼容性;新增 AMD 旧缓存清理、组件后端校验和运行时缺失提示。
  • AI 继续作为可选下载,普通手势无需安装 AI。个人样本和模型不会随组件卸载而删除。

安装与升级

主程序 MSI / 便携版为 x64,需要 .NET 10 Desktop Runtime 和 Windows App Runtime 2.x(至少 2.3.1.0)。AI 组件提供 x64 / ARM64 两种架构。

已有旧 AI 组件会继续使用旧组件,不会自动变小。要使用轻量版,请在新主程序选项中卸载旧组件,选择 CPU,再下载或导入对应的 cpu ZIP;安装后按需重新开启后台学习和 AI 否决。需要硬件加速时选择 NPU / GPU / CPU 并使用 hardware ZIP,两种包选一个即可。

ARM64 AI 引擎需要 ARM64 .NET 10 Runtime,即使前端使用 x64 模拟运行。商店版自带的运行时不代表系统已安装共享运行时。NPU、ARM64 和商店安装流程尚未实机验证。Microsoft Store 与 WinGet 的审核及同步进度独立。

English

  • Default to a lightweight CPU AI component retaining local collection, training, scoring, corrections and AI veto: 0.78 MiB download / 3.15 MiB unpacked on x64.
  • Keep an optional NPU / GPU / CPU component: 24.49 MiB download / 70.55 MiB unpacked on x64, excluding vendor providers, drivers and caches. Inference selects one backend in NPU → GPU → CPU order; NPU and GPU do not run jointly.
  • Both AI variants use the matching architecture of the shared .NET 10 Runtime. Preserve automatic resume for installed 18.3.1 AI components, and add bounded AMD cache cleanup and backend validation.
  • AI remains optional. Switching variants preserves personal samples and models.

The x64 MSI/portable app requires .NET 10 Desktop Runtime and Windows App Runtime 2.x (at least 2.3.1.0). To reduce an existing AI installation, uninstall its component in the updated app, select CPU and download/import the matching CPU ZIP, then re-enable learning/veto as desired. Use the hardware ZIP instead when acceleration is wanted. ARM64 engines require ARM64 .NET 10 Runtime. NVIDIA GPU inference was tested; NPU, ARM64 hardware and Store installation were not device-tested. Store and WinGet availability depends on their independent review/synchronization.