fix bugs: output of splited fc is wrong (#38724) (#38744)

python/paddle/distributed/collective.py

@@ -1176,6 +1176,7 @@ def _parallel_linear(x,
             inputs={'X': linear_out},
             outputs={'Out': out},
             attrs={
+                'rank': inner_rank,
                 'ring_id': ring_id,
                 'nranks': nranks,
                 'use_calc_stream': True,