update mp (#33194)

* update mp

python/paddle/distributed/collective.py

@@ -1009,16 +1009,18 @@ def _parallel_linear(x,
             name=name)
 
     linear_out = linear(x)
-    startup_block = paddle.static.default_startup_program().global_block()
-    main_block = paddle.static.default_main_program().global_block()
-    startup_block.vars[linear.weight.name].is_distributed = True
-    main_block.vars[linear.weight.name].is_distributed = True
+    startup_block = paddle.static.default_startup_program().current_block()
+    main_block = paddle.static.default_main_program().current_block()
+    startup_block._find_var_recursive(linear.weight.name).is_distributed = True
+    main_block._find_var_recursive(linear.weight.name).is_distributed = True
+
     # set is_distributed for splited bias
     # if a linear layer is splited by row, each rank would hold a complete bias and they should be the same in each rank.
     # if a linear layer is splited by col, the bias would also be split into each rank as its weight
     if axis == 1 and linear._bias_attr != False:
-        startup_block.vars[linear.bias.name].is_distributed = True
-        main_block.vars[linear.bias.name].is_distributed = True
+        startup_block._find_var_recursive(
+            linear.bias.name).is_distributed = True
+        main_block._find_var_recursive(linear.bias.name).is_distributed = True
 
     if not gather_out: return linear_out