Skip to content

ANNOUNCEMENT.md

Latest commit

 

History

History
24 lines (18 loc) · 2.85 KB

ANNOUNCEMENT.md

File metadata and controls

24 lines (18 loc) · 2.85 KB

NEW Announcement 3/28/25: THE MERGE IS COMING

Image

Here's how DiffTrainer is handling it:

  • Main(currently single-dict) will get an update that redirects the update functions to a new branch, singledict-archive, and an archived copy of DiffSinger
  • After a few weeks to allow these changes to spread, the current main branch will be replaced with the multi-dict branch
  • When this happens, the multi-dict branch will get an update redirecting the update functions to the main branch, and the current multi-dict branch will be abandoned
  • Single-dict is highly unlikely to receive further updates. Do multi-dict. It's good.

OLD Announcement 2/17/25: Auto-update error ( 中文（正體） )

Versions 0.2.2 through 0.2.9 will no longer correctly trigger the auto-update. To restore this feature, edit difftrainer.py in your choice of text editor and change version: 0.2.x to version: 0.2.0.

OLD Announcement: Major pending changes

Version 0.1.18 will be the last version to support the existing installer and launcher method. Due to directly conflicting package requirements(SOME/training requiring Torch > 2.0, ONNX exporting requiring Torch < 2.0), Conda or self-management of environments will be required. With this in mind, the install and launch .bat files will be replaced with the following steps:

  • (if needed) run conda_installer.bat(formerly python_installer.bat) to quickly install a copy of Miniconda3. The resulting program is not exclusive to DiffTrainer and can be used to manage requirements for other Python 3.x projects. Existing Anaconda/Miniconda installs will be detected by the following setup.
  • run setup.bat to automatically create and configure two environments, DifftrainerA and DifftrainerB. During this process, the torchdrop scripts automatically detect any installed CUDA version and selects appropriate versions of Torch for CPU or GPU.
    • to manually configure environments, requirement files are located in /assets/. Run both torchdrop scripts in their respective environments to detect the required Torch version and complete dependency installation.
  • Use Update Tools on the main tab of DiffTrainer to download the
  • For preprocessing, binarization, and training, use run_guiA.bat to launch. Use run_guiB.bat when exporting ONNX files.

Due to the extensive changes, it is recommended to remove all older versions of DiffTrainer and accompanying files before configuring DiffTrainer 0.2.0. If python_installer.bat was used to install a dedicated copy of Python, use the uninstaller program in the /python folder.

Users are encouraged to manually upgrade to version 0.2.0 in advance of its release on the main branch by downloading the SOME branch.