Remove CUDA PyTorch (Docker related problems)

[MRD2F]

Because your project is a speech-to-text API, you do NOT need CUDA unless you are training big models.

✅ Use CPU torch wheels
They download quickly and work everywhere.

🚀 What you should do RIGHT NOW
Step 1: Remove CUDA PyTorch (most important)

Edit pyproject.toml:

dependencies = [
"torch==2.9.1+cpu",
...
]

Step 2: Now add this section to the bottom of your pyproject.toml:

[tool.uv.index]
default = "https://pypi.org/simple"

[tool.uv.package-index."pytorch-cpu"]
url = "https://download.pytorch.org/whl/cpu"

[tool.uv.dependency-sources]
torch = "pytorch-cpu"

This means:

all packages install from PyPI

only torch installs from the PyTorch CPU index

Step 3: Run

uv sync