[AutoParallel] fix fp16 for subblock (PaddlePaddle#47189)

* [AutoParallel] fix fp16 for subblock

* fix engine

* fix comment

python/paddle/distributed/auto_parallel/engine.py

@@ -240,8 +240,8 @@ def _infer_item_spec(item, name, batch_size, specs):
                 else:
                     specs.append(spec.batch(batch_size))
             elif isinstance(item, (Variable, core.VarBase, core.eager.Tensor)):
-                _adjust_item_spec(num_shards, spec)
                 spec = InputSpec.from_tensor(item, name)
+                _adjust_item_spec(num_shards, spec)
                 if batch_size is None:
                     specs.append(spec)
                 else:
@@ -1508,10 +1508,10 @@ def load(self, path, strict=True, load_optimizer=True):
             strict (bool, optional): Whether to skip the loading of mismatch
                 parameter or raise an error when mismatch happens (not found
                 the parameter in file storing model states of or receives a
-                mismatch shape). Default: False.
+                mismatch shape). Default: True.
             load_optimizer (bool, optional): If True, the stored optimizer
                 states is restored. Otherwise, the optimizer states is initialized
-                from scratch. Default: False.
+                from scratch. Default: True.
 
         Returns:
             None

python/paddle/distributed/passes/auto_parallel_fp16.py

@@ -181,7 +181,8 @@ def set_var_to_fp16(self, var_name, block):
         try:
             var = block.var(var_name)
         except ValueError as e:
-            var = self.program.global_block().var(var_name)
+            var = block._var_recursive(var_name)
+            # var = self.program.global_block().var(var_name)
 
         # NOTE(JZ-LIANG) "array_" is a hack to adopt for ernie3.0 inference, since there is
         # a trick which make the LOD_TENSOR_ARRAY to the float32 in while block to reset the LOD_TENSOR_ARRAY