The dgl.optim.SparseGradOptimizer class and the NodeEmbedding class it manages are currently missing…
The dgl.optim.SparseGradOptimizer class and the NodeEmbedding class it manages are currently missing…: a task in LegoFlow-SWE (Harbor dataset). These methods must work in both single‑process and distributed (multi‑GPU/multi‑process) environments, coordinating via the existing…
The task
These methods must work in both single‑process and distributed (multi‑GPU/multi‑process) environments, coordinating via the existing TCP‑store/communicator infrastructure of `NodeEmbedding`. The `SparseAdagrad` optimizer should also be supported through the same `_all_get_optm_state`/`_all_set_optm_state` mechanisms.
Part of Lego-X/LegoFlow-SWE.