Installation and environments#

How do I verify my installation?#

python -c "import lazyslide as zs; print(zs.__version__)"

Then run a minimal, model-free smoke test:

import lazyslide as zs

wsi = zs.datasets.sample(with_data=False)
zs.pp.find_tissues(wsi, level=-1)
print(len(wsi.shapes["tissues"]))

Which slide reader should I install?#

TiffSlide and OpenSlide cover many bright-field WSI formats. fastslide provides a high-performance native reader; Bio-Formats supports broad microscopy coverage; cuCIM supports GPU-accelerated I/O; pyisyntax handles Philips iSyntax; and pylibCZIrw handles Zeiss CZI. Reader installation details are on the Installation page.

Test the selected reader against representative files from your scanner. A filename extension alone does not guarantee that every vendor variant is supported.

How do I access a gated Hugging Face model?#

Request access on the model’s Hugging Face page, create a read token, and authenticate once:

hf auth login

Do not place tokens in notebooks, command history, or committed configuration. Confirm access by instantiating the model before launching a long slide job.

How do I run models on an offline compute node?#

Download the weights on a networked machine or login node:

hf download OWNER/MODEL

Copy or mount the same Hugging Face cache on the compute node, then set offline mode before importing model libraries:

export HF_HUB_OFFLINE=1

Set HF_HOME consistently if the cache is not under its default location. See the Model Zoo for the complete offline workflow.

How do I install the development documentation environment?#

From a repository checkout:

uv sync --group docs --group tutorials
uv run task doc-build

Contributor setup is documented in Development setup.