[BugFix] fix incorrect name when fetch data in sparse optim (#3808)

python/dgl/distributed/nn/pytorch/sparse_emb.py

@@ -130,6 +130,17 @@ class DistEmbedding:
         """
         return self._tensor.tensor_name
 
+    @property
+    def data_name(self):
+        """Return the data name of the embeddings
+
+        Returns
+        -------
+        str
+            The data name of the embeddings
+        """
+        return self._tensor._name
+
     @property
     def kvstore(self):
         """Return the kvstore client

python/dgl/distributed/optim/pytorch/sparse_optim.py

@@ -74,7 +74,7 @@ class DistSparseGradOptimizer(abc.ABC):
                 # will send grad to each corresponding trainer
                 if self._world_size > 1:
                     # get idx split from kvstore
-                    idx_split = kvstore.get_partid(name, idics)
+                    idx_split = kvstore.get_partid(emb.data_name, idics)
                     idx_split_size = []
                     idics_list = []
                     grad_list = []