cugraph
cugraph copied to clipboard
cuGraph-DGL and WholeGraph Performance Testing with Feature Store Performance Improvements
Large-scale cuGraph-DGL performance testing scripts. Also changes the DGL and PyG scripts to evaluate on all ranks and reuse the test samples, and adds support for benchmarking cuGraph-DGL/cuGraph-PyG with WholeGraph.
Updates cuGraph.gnn.FeatureStore
and cuGraph-PyG
for increased performance:
- Supporting passing in a WG embedding directly to cugraph.gnn.FeatureStore
- Simplifying how cuGraph-PyG handles filtering and using a cache to prevent repeatedly copying data between the device and host
- Fix bug in cugraph.gnn.FeatureStore where indexing with a gpu tensor would raise an exception, especially with WG
- Add a function to cugraph.gnn.FeatureStore to check where data is stored, which is used by cuGraph-PyG to prevent unnecessary d2h and h2d copies
Merge after #3584
This pull request requires additional validation before any workflows can run on NVIDIA's runners.
Pull request vetters can view their responsibilities here.
Contributors can view more details about this message here.
/ok to test
/ok to test
/ok to test
/ok to test
/merge
/ok to test
/merge
/ok to test
/merge
/ok to test
/ok to test
/ok to test
/ok to test
/ok to test
/ok to test
/ok to test