v3.0.1

Huanshere/VideoLingov3.0.1Feb 28, 2026by Huanshere

AI Summary

Adds RTX 50 series GPU support and upgrades core dependencies including PyTorch and whisperX.

Key Highlights

  • RTX 50 series (Blackwell) GPU support with sm_100 kernels.
  • Upgraded to PyTorch 2.8.0 and whisperX 3.8.1.
  • Fixed JSON encoding for Chinese characters in prompts.

New Features

  • RTX 50 series (Blackwell) GPU support
  • PyTorch and whisperX dependency upgrades

Full Release Notes

Changelog - Version 3.0.1
This update adds RTX 50 series (Blackwell) GPU support, upgrades core dependencies, and fixes several deprecation warnings and stability issues.

🚀 New Features
- RTX 50 series (Blackwell) GPU support: installer now uses `nvidia-smi` to detect driver CUDA version and automatically selects cu129 wheels with sm_100 kernels for Blackwell GPUs.
- Upgraded to PyTorch 2.8.0 / whisperX 3.8.1 with improved CUDA wheel auto-selection (cu129 / cu128 / cu126).

🐛 Bug Fixes
- Fixed ffmpeg fallback to WAV encoding when libmp3lame is unavailable (#531).
- Fixed JSON encoding for Chinese characters in prompts.
- Fixed term extraction prompt to handle proper noun translation correctly.
- Fixed Streamlit audio deadlock caused by `librosa.load()` in non-main thread.

🔧 Improvements
- Replaced deprecated `pynvml` package with official `nvidia-ml-py`.
- Replaced deprecated Streamlit `use_container_width` API with new `width` parameter.
- Resolved dependency conflicts for PyTorch / torchaudio upgrade.
- Updated installation guide and Docker docs with cu129 support, model recommendations, and pitfall warnings.

---

更新日志 - 版本 3.0.1
本次更新新增了 RTX 50 系列(Blackwell)显卡支持,升级核心依赖,并修复了多个弃用警告和稳定性问题。

🚀 新功能
- RTX 50 系列(Blackwell)显卡支持:安装脚本现通过 `nvidia-smi` 检测驱动 CUDA 版本,自动为 Blackwell 显卡选择包含 sm_100 内核的 cu129 轮子。
- 升级至 PyTorch 2.8.0 / whisperX 3.8.1,改进 CUDA 轮子自动选择(cu129 / cu128 / cu126)。

🐛 问题修复
- 修复 ffmpeg 缺少 libmp3lame 时回退到 WAV 编码的问题(#531)。
- 修复提示词中中文字符的 JSON 编码问题。
- 修复术语提取 prompt 中专有名词翻译的处理。
- 修复 Streamlit 中 `librosa.load()` 在非主线程导致的音频死锁。

🔧 改进
- 将已弃用的 `pynvml` 替换为官方 `nvidia-ml-py`。
- 将已弃用的 Streamlit `use_container_width` API 替换为新的 `width` 参数。
- 解决 PyTorch / torchaudio 升级的依赖冲突。
- 更新安装指南和 Docker 文档:cu129 支持说明、模型推荐、踩坑提示。