v1.5.0

Version 1.5.0

v1.4.11

fix: CUDA版本改为运行时下载，避免打包时体积过大导致NSIS失败

- 移除requirements_win_cuda.txt中的torch和torchaudio依赖
- 首次启动时自动检测CUDA模式并提醒用户下载PyTorch CUDA
- 优先使用南京大学镜像源下载，失败后自动切换到官方源
- 在设置中提供一键下载PyTorch CUDA的功能

v1.4.10

fix: 修复Windows版本requirements文件中包名合并的问题

v1.4.8

Release v1.4.8: 优化PyTorch CUDA下载，优先使用国内镜像源

v1.4.6

Release v1.4.6: 修复CUDA版本构建失败，改用zip打包

v1.4.5

Release v1.4.5: 优化分块处理和日志输出，移除torchaudio依赖

v1.4.4

Fix: Add missing dependency latex2mathml for LaTeX rendering

v1.4.3

Fix: Use asyncio.call_soon_threadsafe for UI updates

v1.3.7

fix: mlx_whisper API and model path resolution

- Fix mlx_whisper API: use path_or_hf_repo instead of deprecated model param
- Remove ui.notify calls from background threads (causes RuntimeError)
- For MLX backend, return repo_id directly since mlx_whisper handles cache
- Update complexity option labels with TOC generation rules

v1.3.6

fix(cuda): remove torchaudio to reduce binary size below 2GB NSIS limit

- Remove torchaudio from --collect-all and --collect-binaries
- faster-whisper uses ctranslate2, not torchaudio
- CUDA binary was 2.5GB, exceeding NSIS 2GB limit