notebook-graph

Version	2026.144.1443
Repo	superproject

Standalone marimo notebook exercising the cuGraph + cuML + PyG + graphistry GPU libraries Provisions a self-contained marimo notebook into /workspace/notebooks/ at deploy time. The notebook demonstrates the four new GPU library families added to the versa env in the 2026-05 cutover:

• cuGraph (RAPIDS graph analytics) — PageRank via the nx-cugraph backend pattern (cugraph.datasets was removed in 26.4, so nx.pagerank(G, backend=“cugraph”) is the idiomatic accelerator path).
• cuML (RAPIDS ML) — KMeans clustering on synthetic blobs, end-to-end on GPU.
• PyTorch Geometric — GCNConv forward pass on cuda:0, with torch_scatter / torch_sparse / torch_cluster providing the accelerated scatter/gather kernels.
• graphistry — GPU-accelerated graph visualization client; the notebook builds an edges DataFrame and shows how to construct a plotter without requiring a Graphistry server account.

Data-only layer — no packages, no services, no baked DAGs. The notebook reads no env vars (every cell is self-contained synthetic data) so it runs cleanly even when the OSM DAG hasn’t produced /workspace/tiles/work/ yet.

Acceptance plan

This candy’s plan: — the runnable spec charly check executes against a live deployment. check: steps are idempotent probes; run: steps change state.

Intent	Step
check	the notebooks subdirectory is provisioned in the workspace
check	the gpu-libraries-demo notebook is present
check	file=${VOLUME_CONTAINER_PATH:workspace}/notebooks/gpu-libraries-demo.py