fix python2 (#660)

paddleslim/core/dygraph.py

@@ -61,13 +61,13 @@ def extract_vars(inputs):
                 vars.append(_value)
             else:
                 _logger.warn(
-                    f"Variable is excepted, but get an element with type({type(_value)}) from inputs whose type is dict. And the key of element is {_key}."
+                    "Variable is excepted, but get an element with type({}) from inputs whose type is dict. And the key of element is {}.".format(type(_value), _key)
                 )
     elif isinstance(inputs, (tuple, list)):
         for _value in inputs:
             vars.extend(extract_vars(_value))
     if len(vars) == 0:
-        _logger.warn(f"Extract none variables from inputs.")
+        _logger.warn("Extract none variables from inputs.")
     return vars
 
 

paddleslim/dygraph/prune/filter_pruner.py

@@ -320,7 +320,7 @@ class FilterPruner(Pruner):
         """
         if var_name in self.skip_vars:
             _logger.warn(
-                f"{var_name} is skiped beacause it is not support for pruning derectly."
+                "{} is skiped beacause it is not support for pruning derectly.".format(var_name)
             )
             return
         if isinstance(pruned_dims, int):
@@ -340,7 +340,7 @@ class FilterPruner(Pruner):
                             'var': param,
                             'value': np.array(param.value().get_tensor())
                         })
-                    _logger.debug(f"set value of {param.name} into group")
+                    _logger.debug("set value of {} into group".format(param.name))
 
         mask = self.cal_mask(var_name, pruned_ratio, group_dict)
         for _name in group_dict:
@@ -356,7 +356,7 @@ class FilterPruner(Pruner):
                     src_mask = self._transform_mask(src_mask, trans)
                 current_mask = src_mask
                 assert len(current_mask) == var_shape[dims[
-                    0]], f"The length of current_mask must be equal to the size of dimension to be pruned on. But get: len(current_mask): {len(current_mask)}; var_shape: {var_shape}; dims: {dims}; var name: {_name}; len(mask): {len(mask)}"
+                    0]], "The length of current_mask must be equal to the size of dimension to be pruned on. But get: len(current_mask): {}; var_shape: {}; dims: {}; var name: {}; len(mask): {}".format(len(current_mask), var_shape, dims, _name, len(mask))
                 plan.add(_name, PruningMask(dims, current_mask, pruned_ratio))
         if apply == "lazy":
             plan.apply(self.model, lazy=True)

paddleslim/prune/group_param.py

@@ -60,7 +60,7 @@ def collect_convs(params, graph, visited={}):
         param = graph.var(_param)
         if param is None:
             _logger.warning(
-                f"Cann't found relative variables of {_param} because {_param} is not in target program or model. Please make sure {_param} is in your program if you are using static API of PaddlePaddle. And make sure your model in correctly mode and contains {_param} if you are using dynamic API of PaddlePaddle."
+                "Cann't found relative variables of {} because {} is not in target program or model. Please make sure {} is in your program if you are using static API of PaddlePaddle. And make sure your model in correctly mode and contains {} if you are using dynamic API of PaddlePaddle.".format(_param, _param, _param, _param)
             )
             groups.append([])
             continue

paddleslim/prune/prune_walker.py

@@ -95,7 +95,7 @@ class PruneWorker(object):
         _logger.debug("\nfrom: {}\nto: {}\npruned_axis: {}; var: {}".format(
             self.op, op, pruned_axis, var.name()))
         _logger.debug(
-            f"visit {op.type()} by var [{var.name()}] on axis [{pruned_axis}];\t visited={self.visited}\n"
+            "visit {} by var [{}] on axis [{}];\t visited={}\n".format(op.type(), var.name(), pruned_axis, self.visited)
         )
         walker = cls(op, pruned_params=self.pruned_params, visited=self.visited)
         walker.prune(var, pruned_axis, pruned_idx)
@@ -123,7 +123,7 @@ class conv2d(PruneWorker):
     def _prune(self, var, pruned_axis, pruned_idx):
 
         if self._is_depthwise_conv(self.op):
-            _logger.debug(f"Meet conv2d who is depthwise conv2d actually.")
+            _logger.debug("Meet conv2d who is depthwise conv2d actually.")
             walker = depthwise_conv2d(
                 self.op, self.pruned_params, visited=self.visited)
             walker._prune(var, pruned_axis, pruned_idx)

paddleslim/prune/pruner.py

@@ -144,7 +144,7 @@ class Pruner():
                                 op.attr("groups") * new_shape[pruned_axis] /
                                 origin_shape[pruned_axis])
                             _logger.debug(
-                                f"change groups of conv({param.name()}) from {op.attr('groups')} to {new_groups}; origin_shape: {origin_shape}; new_shape: {new_shape}"
+                                "change groups of conv({}) from {} to {}; origin_shape: {}; new_shape: {}".format(param.name(), op.attr('groups'), new_groups, origin_shape, new_shape)
                             )
                             op.set_attr("groups", new_groups)