Installation¶
Requirements¶
- Python ≥ 3.13 (declared in
pyproject.tomland.python-version) - uv — recommended package and environment manager
- Git — for cloning the repository and CLARIN.SI dataset access
- HuggingFace account — required for gated datasets (e.g.,
FineWeb-2); see HuggingFace Authentication - Disk space — pretraining corpora total tens of GB; plan accordingly
- GPU (optional) — required for tokenizer/model training; CPU sufficient for data preparation
Clone and install¶
Clone the repository and create the virtual environment via uv:
This creates .venv/ and installs both runtime and dev dependencies pinned in uv.lock. Activate the environment for ad-hoc commands:
Or prefix individual commands with uv run to skip activation.
Optional extras¶
The base uv sync only pulls runtime essentials. Install extras as needed:
| Extra | Command | Purpose |
|---|---|---|
dev | uv sync --extra dev | Test runner (pytest) and linter (ruff) for contributors |
curate | uv sync --extra curate | datatrove pipeline for the pretraining corpus |
tokenize | uv sync --extra tokenize | Tokenizer training + morphological evaluation (see Tokenizers) |
docs | uv sync --extra docs | MkDocs Material toolchain for building this documentation site |
notebook | uv sync --extra notebook | marimo notebooks + plotly/kaleido for experiment reporting and visualization |
Multiple extras can be combined:
Next steps¶
- Authenticate with HuggingFace: HuggingFace Authentication
- Get oriented in the codebase: Project Structure
- Start running the pipeline: User Guide