create-ml-stack 0.0.1

Better-T-Stack for ML — scaffold end-to-end runnable ML projects

create-ml-stack

Better-T-Stack for ML — scaffold end-to-end runnable machine learning projects in seconds.

uvx create-ml-stack my-ml-app
# or: pipx run create-ml-stack my-ml-app

Choose a framework (PyTorch, JAX, scikit-learn, HF Transformers), tracking, data versioning, config, and serving — then get a project that trains, tracks, and serves on the first try.

Quick start

uvx create-ml-stack my-ml-app
cd my-ml-app
uv sync --python 3.11   # or: conda env create -f environment.yml / pip install -r requirements.txt
uv run train --fast     # tiny seed data, <90s on CPU
uv run serve --dry-run  # load model without binding a port
uv run serve            # start local serving helper

Features

• Multi-framework: PyTorch, JAX, scikit-learn, Hugging Face Transformers
• Integrations: W&B / MLflow, DVC, Hydra (+ Optuna), BentoML / Modal / HF Spaces
• Package managers: uv (recommended), conda, or pip
• Layouts: monorepo (packages/) or single-package (src/)
• Hybrid compatibility matrix: curated fast path + on-the-fly resolver fallback
• Offline-first: vendored tiny MNIST/digits shards for happy-path training

Non-interactive flags

create-ml-stack my-app \
  --framework torch \
  --tracking wandb \
  --data dvc \
  --config hydra \
  --serving bentoml \
  --pm uv \
  --layout monorepo \
  --yes

Development

uv sync --extra dev
uv run pytest -q
uv run ruff check .
uv build   # verify wheel before release

See docs/publishing.md for PyPI release steps.

License

MIT