first commit

starrygl/sample/part_utils/transformer_from_speed.py

@@ -136,7 +136,7 @@ def load_from_shared_node_partition(data,node_i,shared_node,sample_add_rev = Tru
     test_mask = (torch.from_numpy(np.array(df.ext_roll.values)) == 2)   
     test_mask = (torch.from_numpy(np.array(df.ext_roll.values)) == 2)   
     num_node = max(src.max().item(),dst.max().item())+1 
     num_node = max(src.max().item(),dst.max().item())+1 
     dim_feats = [0, 0, 0, 0, 0, 0, 0, 0]
     dim_feats = [0, 0, 0, 0, 0, 0, 0, 0]
-    if dist.get_rank() == 0:
+    if ctx.local_rank == 0:
         _node_feats, _edge_feats = load_feat(data, num_node, src.shape[0], 172, 172)
         _node_feats, _edge_feats = load_feat(data, num_node, src.shape[0], 172, 172)
         if _node_feats is not None:
         if _node_feats is not None:
             dim_feats[0] = _node_feats.shape[0]
             dim_feats[0] = _node_feats.shape[0]
@@ -165,7 +165,7 @@ def load_from_shared_node_partition(data,node_i,shared_node,sample_add_rev = Tru
     torch.distributed.broadcast_object_list(dim_feats, src=0)
     torch.distributed.broadcast_object_list(dim_feats, src=0)
     print('dist rank is {} after node feats defination:'.format(torch.distributed.get_rank()))
     print('dist rank is {} after node feats defination:'.format(torch.distributed.get_rank()))
     print()
     print()
-    if dist.get_rank() > 0:
+    if ctx.local_rank > 0:
         node_feats = None
         node_feats = None
         edge_feats = None
         edge_feats = None
         if dim_feats[6] == 1:
         if dim_feats[6] == 1: