pycu-cuda 0.1.0

PyCU CUDA package - minimal CUDA detection and stubs

PyCU

Minimal PyCU: a small pure-Python scaffold showing a CPU tensor implementation and a lightweight CUDA detection stub. This repo is a starting point — a full PyTorch replacement is out of scope for a single scaffold and requires native bindings and drivers.

Quick usage

1. Install locally for development:

python -m pip install --upgrade build
python -m build
pip install dist/PyCU-0.1.0-py3-none-any.whl

2. Use the library:

from pycu import tensor, cuda_available, get_cuda_version

print('cuda available:', cuda_available())
print('cuda version:', get_cuda_version())

T = tensor([[1,2],[3,4]], device='cpu')
S = tensor([[5,6],[7,8]], device='cpu')
print('matmul:', T.matmul(S).tolist())

Publishing to PyPI

• I cannot use or upload with credentials you provided. Do NOT share tokens.
• To publish yourself, build the package and upload with twine:

python -m pip install --upgrade build twine
python -m build
python -m twine upload dist/* -u __token__ -p <YOUR_TOKEN_HERE>

If you want separate PyPI package names (PyCU-CPU and PyCU-CUDA), change the name field in pyproject.toml before building.

Publishing pycu-cpu and pycu-cuda

This repo can be published as two separate PyPI packages while keeping the import name as pycu (the distribution name and import package name are independent). Two sample pyproject files are provided: pyproject_cpu.toml and pyproject_cuda.toml.

Build and upload pycu-cpu:

# switch to cpu pyproject
copy pyproject_cpu.toml pyproject.toml /Y
python -m build
# set env vars securely before upload
$env:TWINE_USERNAME = "__token__"
$env:TWINE_PASSWORD = "<YOUR_TOKEN>"
python -m twine upload --repository pypi --verbose dist/*

Build and upload pycu-cuda:

# switch to cuda pyproject
copy pyproject_cuda.toml pyproject.toml /Y
python -m build
# set TWINE_USERNAME/TWINE_PASSWORD as above
python -m twine upload --repository pypi --verbose dist/*

Important: create a fresh PyPI API token for each upload (do not reuse the exposed token), use __token__ as the username and the token string as the password, and never paste tokens into public chat.