[FEA] Migrate All PyTorch-Dependent Code to cuGraph-GNN #4822
Comments
|
Would |
|
Right now it's used by cuGraph-PyG and cuGraph-DGL. The long-term goal is to integrate with other frameworks, including pure PyTorch. But I suppose we might still be able to support that within cuGraph-PyG. So then only cuGraph-DGL would be the other dependency, and we have been debating what to do about that package. |
|
Also there is no |
Follow up to #4935 There, in #4935 (comment), we discussed the that GNN packages shouldn't need to be installed in docs builds any more, as no docs in this repo (including notebooks) require them. This PR limits dependencies on the GNN packages to only the places they're needed. ### `libwholegraph` / `pylibwholegraph` ```shell git grep -i -E 'wholegraph' ``` Optional runtime dependency of `cugraph`: https://github.com/rapidsai/cugraph/blob/2873ff91945c4944568ffd1aa035f6bba17746a0/python/cugraph/cugraph/gnn/feature_storage/feat_storage.py#L23 And optional test-time dependency of `cugraph`: https://github.com/rapidsai/cugraph/blob/2873ff91945c4944568ffd1aa035f6bba17746a0/python/cugraph/cugraph/tests/data_store/test_gnn_feat_storage_wholegraph.py#L24-L25 But not used in any docs. **Changes:** Removed from `docs` environment. ### `cugraph-dgl` ```shell git grep -i -E 'cugraph.*dgl' ``` Not used anywhere in this repo. **Changes:** Removed all remaining references. ### `cugraph-pyg` ```shell git grep -i -E 'cugraph.*pyg' ``` Only used as an optional import in `cugraph-service-client`'s tests (which are not run in CI for wheels). https://github.com/rapidsai/cugraph/blob/2873ff91945c4944568ffd1aa035f6bba17746a0/python/cugraph-service/tests/test_remote_graph.py#L662 **Changes:** Removed from the `docs` environment, added a `[test]` extra to `cugraph-service-client` including this. ## Notes for Reviewers Related to these issues about moving more GNN stuff out of this repo: * #4822 * #4407 Authors: - James Lamb (https://github.com/jameslamb) Approvers: - Chuck Hastings (https://github.com/ChuckHastings) - Mike Sarahan (https://github.com/msarahan) - Bradley Dice (https://github.com/bdice) URL: #4947
We continue to have issues with PyTorch in this repository's CI. In addition, the code that depends on PyTorch is really only used by cuGraph-GNN and does not really fit the mission of this repository. As discussed over the past few months, we want to migrate as much GNN code as possible to the
cugraph-gnnrepository.There are three key pieces of code affected:
FeatureStoreclass which is about to be deprecated (in release 25.02)BulkSamplerclass which is also about to be deprecated (in release 25.02)DistSamplerclass, the replacement forBulkSampler, which is a fundamental piece of our GNN infrastructure.There is also going to be some additional code in the very near future supporting GNN use cases related to GraphRAG, graph databases, and other frameworks (beyond DGL and PyG). This would also better fit within
cugraph-gnn.We propose creating a new package,
pylibcugraphgnn, which will contain the bulk sampling code, as well as any other framework-agnostic code and/or thin wrappers around our C++ code for GNN operations. This package will presumably launch with release 25.04.The text was updated successfully, but these errors were encountered: