Some notes at windows 10 installation

[ErunamoJAZZ]

A simple log of my little journey installing this in my osu pc:

1. Installed Python from the Windows Store (first time trying this method lol - Warning: notice the item #5)
2. When .venv\Scripts\Activate.ps1 in PowerShell, I got an error (about_Execution_Policies).
   a. Check with Get-ExecutionPolicy returning Restricted
   b. And fix for the current session with Set-ExecutionPolicy Unrestricted -Scope Process.
   c. Now Get-ExecutionPolicy should return Unrestricted.
3. On pip install -r requirements.txt failed:
   a. WARNING: Failed to activate VS environment: Could not find C:\Program Files (x86)\Microsoft Visual Studio\Installer\vswhere.exe <- it's because no C++ compiler was found, visit https://visualstudio.microsoft.com/en/visual-cpp-build-tools/ and download the tools.
   b. Be sure of installing "Desktop development with C++" (it's like 6GB)
4. Mess with PIP version: Since pip 24.1 (current version is 24.3), some dependencies with wrong listed versions will fail on installing. This will require to change some dependencies?, not sure, but anyway...
   a. A faster solution would be to downgrade pip to 24.0 python.exe -m pip install pip==24.0
   b. Check version with pip -V
5. This still failing with a weird error: ERROR: Cannot install -r requirements.txt (line 15) because these package versions have conflicting dependencies.
   a. That was weird, but the final reason was because Windows Store suggested me to install the last version of python 3.13, but this project require Python 3.10 lol
   b. After installing the right version of python, I just fixed .venv\pyvenv.cfg with the right python version/path, and re-run the installation, and this finally worked, yey!
6. At first run with only audio, I got a CUDA error: AssertionError: Torch not compiled with CUDA enabled, so I download it from https://developer.nvidia.com/cuda-downloads and also re install pytorch with cuda support with pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124 (for the future, check the current command/CudaVersion at "install pytorch" section in https://pytorch.org/)
7. Noticed a warning, Couldn't find ffmpeg or avconv - defaulting to ffmpeg, but may not work. So I downloaded this from https://www.gyan.dev/ffmpeg/builds/ and placed the exes in the folder.

Finally this worked with no more issues 💯

HOWEVER!, output was in the Limbo 🙃 ahahaha.
Making the default output_path a '.' instead an empty string, works.

Result was interesting, but not great for my example song :P

Great job!

[OliBomby]

Nice write-up of your journey. I could improve the documentation and default settings to make things easier.

I also noticed you ran out of dedicated VRAM in your screenshot and timing generation took a long while. If you lower the max_batch_size argument it should fix that.

[ErunamoJAZZ]

Default max_batch_size was 16, changing it to 8 seems to run faster in my 8GB gpu :P

[OliBomby]

I'll look into making it auto-detect the available vram and scaling max_batch_size accordingly.