Install¶
cognarium needs Python 3.11 or later and PyTorch 2.9 or later.
uv add cognarium
or, with pip:
pip install cognarium
The base install includes everything cg.load("gpt2") and the dataset registry need,
such as Transformers, the Hugging Face Hub client and pooch.
From source¶
git clone https://github.com/IParraMartin/cognarium
cd cognarium
uv sync --extra dev
uv run pytest
Extras¶
| Extra | Adds |
|---|---|
dev |
The test suite's tools, and the libraries its oracle tests compare against: SciPy, scikit-learn, dit and rsatoolbox |
docs |
MkDocs Material and mkdocstrings, to build this site |
Devices¶
device="auto" picks CUDA, then Apple MPS, then the CPU. dtype="auto" picks a
precision the device supports. Metrics that need float64, or that have known problems
on MPS, always compute on the CPU, whatever device the model uses.
import cognarium as cg
subject = cg.load("gpt2", device="auto")
subject.device