[Dygraph] Tensor Parallelism improve & accuracy alignment

python/paddle/distributed/collective.py

@@ -1008,6 +1008,10 @@ def _parallel_linear(x,
     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
+    # set is_distributed for splited bias
+    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
 
     if not gather_out: return linear_out
 

python/paddle/fluid/contrib/mixed_precision/fp16_lists.py

@@ -145,6 +145,7 @@ def _update_list(self):
     'sign',
     'cast',
     'fused_bn_add_activation',
+    'c_identity',
 }
 
 # The set of ops that don't support fp16 calculation