diff --git a/ppdet/core/workspace.py b/ppdet/core/workspace.py
index b151b15aae706e6033e0618f8b03dc1690948d23..666ca53ad69024439439ddf5e958c61fef44ae02 100644
--- a/ppdet/core/workspace.py
+++ b/ppdet/core/workspace.py
@@ -248,7 +248,7 @@ def create(cls_or_name, **kwargs):
                 if isinstance(target, SchemaDict):
                     kwargs[k] = create(target_key)
                 elif hasattr(target, '__dict__'):  # serialized object
-                    kwargs[k] = new_dict
+                    kwargs[k] = target
             else:
                 raise ValueError("Unsupported injection type:", target_key)
     # prevent modification of global config values of reference types
diff --git a/ppdet/modeling/architecture/meta_arch.py b/ppdet/modeling/architecture/meta_arch.py
index b758e816380876d013989b37e0f394e42974e57e..1a4f4f2591465a3d3920ef886f9bb06a76cd5322 100644
--- a/ppdet/modeling/architecture/meta_arch.py
+++ b/ppdet/modeling/architecture/meta_arch.py
@@ -3,8 +3,8 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from paddle.fluid.dygraph import Layer
-from paddle.fluid.dygraph.base import to_variable
+import paddle
+import paddle.nn as nn
 from ppdet.core.workspace import register
 from ppdet.utils.data_structure import BufferDict
 
@@ -12,7 +12,7 @@ __all__ = ['BaseArch']
 
 
 @register
-class BaseArch(Layer):
+class BaseArch(nn.Layer):
     def __init__(self):
         super(BaseArch, self).__init__()
 
@@ -39,10 +39,10 @@ class BaseArch(Layer):
                 input_v = np.array(input)[np.newaxis, ...]
                 inputs[name].append(input_v)
         for name in input_def:
-            inputs[name] = to_variable(np.concatenate(inputs[name]))
+            inputs[name] = paddle.to_tensor(np.concatenate(inputs[name]))
         return inputs
 
-    def model_arch(self, mode):
+    def model_arch(self):
         raise NotImplementedError("Should implement model_arch method!")
 
     def loss(self, ):
diff --git a/ppdet/modeling/backbone/darknet.py b/ppdet/modeling/backbone/darknet.py
index cce2647237890c95dc1d8e9329b55b7cf653f45e..47d8a077b503c92a867e2eb120d22efcacb98039 100755
--- a/ppdet/modeling/backbone/darknet.py
+++ b/ppdet/modeling/backbone/darknet.py
@@ -1,14 +1,14 @@
-import paddle.fluid as fluid
-from paddle.fluid.dygraph import Layer
-from paddle.fluid.param_attr import ParamAttr
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
 from paddle.fluid.regularizer import L2Decay
-from paddle.fluid.dygraph.nn import Conv2D, BatchNorm
 from ppdet.core.workspace import register, serializable
 
 __all__ = ['DarkNet', 'ConvBNLayer']
 
 
-class ConvBNLayer(Layer):
+class ConvBNLayer(nn.Layer):
     def __init__(self,
                  ch_in,
                  ch_out,
@@ -20,25 +20,22 @@ class ConvBNLayer(Layer):
                  name=None):
         super(ConvBNLayer, self).__init__()
 
-        self.conv = Conv2D(
-            num_channels=ch_in,
-            num_filters=ch_out,
-            filter_size=filter_size,
+        self.conv = nn.Conv2d(
+            in_channels=ch_in,
+            out_channels=ch_out,
+            kernel_size=filter_size,
             stride=stride,
             padding=padding,
             groups=groups,
-            param_attr=ParamAttr(name=name + '.conv.weights'),
-            bias_attr=False,
-            act=None)
+            weight_attr=ParamAttr(name=name + '.conv.weights'),
+            bias_attr=False)
         bn_name = name + '.bn'
-        self.batch_norm = BatchNorm(
-            num_channels=ch_out,
-            param_attr=ParamAttr(
+        self.batch_norm = nn.BatchNorm2d(
+            ch_out,
+            weight_attr=ParamAttr(
                 name=bn_name + '.scale', regularizer=L2Decay(0.)),
             bias_attr=ParamAttr(
-                name=bn_name + '.offset', regularizer=L2Decay(0.)),
-            moving_mean_name=bn_name + '.mean',
-            moving_variance_name=bn_name + '.var')
+                name=bn_name + '.offset', regularizer=L2Decay(0.)))
 
         self.act = act
 
@@ -46,11 +43,11 @@ class ConvBNLayer(Layer):
         out = self.conv(inputs)
         out = self.batch_norm(out)
         if self.act == 'leaky':
-            out = fluid.layers.leaky_relu(x=out, alpha=0.1)
+            out = F.leaky_relu(out, 0.1)
         return out
 
 
-class DownSample(Layer):
+class DownSample(nn.Layer):
     def __init__(self,
                  ch_in,
                  ch_out,
@@ -75,7 +72,7 @@ class DownSample(Layer):
         return out
 
 
-class BasicBlock(Layer):
+class BasicBlock(nn.Layer):
     def __init__(self, ch_in, ch_out, name=None):
         super(BasicBlock, self).__init__()
 
@@ -97,11 +94,11 @@ class BasicBlock(Layer):
     def forward(self, inputs):
         conv1 = self.conv1(inputs)
         conv2 = self.conv2(conv1)
-        out = fluid.layers.elementwise_add(x=inputs, y=conv2, act=None)
+        out = paddle.add(x=inputs, y=conv2)
         return out
 
 
-class Blocks(Layer):
+class Blocks(nn.Layer):
     def __init__(self, ch_in, ch_out, count, name=None):
         super(Blocks, self).__init__()
 
@@ -127,7 +124,7 @@ DarkNet_cfg = {53: ([1, 2, 8, 8, 4])}
 
 @register
 @serializable
-class DarkNet(Layer):
+class DarkNet(nn.Layer):
     def __init__(self,
                  depth=53,
                  freeze_at=-1,
diff --git a/ppdet/modeling/bbox.py b/ppdet/modeling/bbox.py
index fefc3b51efc9da7e9cb4c2bf6907be95e9106140..f60cc65e10478cb5308071d424b5b3ca37643b46 100644
--- a/ppdet/modeling/bbox.py
+++ b/ppdet/modeling/bbox.py
@@ -1,5 +1,8 @@
 import numpy as np
 import paddle.fluid as fluid
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
 from ppdet.core.workspace import register
 
 
@@ -90,9 +93,9 @@ class BBoxPostProcessYOLO(object):
                                           self.num_classes, i)
 
             boxes_list.append(boxes)
-            scores_list.append(fluid.layers.transpose(scores, perm=[0, 2, 1]))
-        yolo_boxes = fluid.layers.concat(boxes_list, axis=1)
-        yolo_scores = fluid.layers.concat(scores_list, axis=2)
+            scores_list.append(paddle.transpose(scores, perm=[0, 2, 1]))
+        yolo_boxes = paddle.concat(boxes_list, axis=1)
+        yolo_scores = paddle.concat(scores_list, axis=2)
         bbox = self.nms(bboxes=yolo_boxes, scores=yolo_scores)
         # TODO: parse the lod of nmsed_bbox
         # default batch size is 1
diff --git a/ppdet/modeling/head/yolo_head.py b/ppdet/modeling/head/yolo_head.py
index cae1f4024e889464d9401d5a0b1d5ca57e1f1b56..42beb69e0cdf168df9a2c34f9df5cf23c3d27b21 100644
--- a/ppdet/modeling/head/yolo_head.py
+++ b/ppdet/modeling/head/yolo_head.py
@@ -1,16 +1,14 @@
 import paddle.fluid as fluid
 import paddle
-from paddle.fluid.dygraph import Layer
-from paddle.fluid.param_attr import ParamAttr
-from paddle.fluid.initializer import Normal
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
 from paddle.fluid.regularizer import L2Decay
-from paddle.fluid.dygraph.nn import Conv2D, BatchNorm
-from paddle.fluid.dygraph import Sequential
 from ppdet.core.workspace import register
 from ..backbone.darknet import ConvBNLayer
 
 
-class YoloDetBlock(Layer):
+class YoloDetBlock(nn.Layer):
     def __init__(self, ch_in, channel, name):
         super(YoloDetBlock, self).__init__()
         self.ch_in = ch_in
@@ -26,7 +24,7 @@ class YoloDetBlock(Layer):
             #['tip', channel, channel * 2, 3],
         ]
 
-        self.conv_module = Sequential()
+        self.conv_module = nn.Sequential()
         for idx, (conv_name, ch_in, ch_out, filter_size,
                   post_name) in enumerate(conv_def):
             self.conv_module.add_sublayer(
@@ -52,7 +50,7 @@ class YoloDetBlock(Layer):
 
 
 @register
-class YOLOFeat(Layer):
+class YOLOFeat(nn.Layer):
     __shared__ = ['num_levels']
 
     def __init__(self, feat_in_list=[1024, 768, 384], num_levels=3):
@@ -88,19 +86,19 @@ class YOLOFeat(Layer):
         yolo_feats = []
         for i, block in enumerate(body_feats):
             if i > 0:
-                block = fluid.layers.concat(input=[route, block], axis=1)
+                block = paddle.concat([route, block], axis=1)
             route, tip = self.yolo_blocks[i](block)
             yolo_feats.append(tip)
 
             if i < self.num_levels - 1:
                 route = self.route_blocks[i](route)
-                route = fluid.layers.resize_nearest(route, scale=2.)
+                route = F.resize_nearest(route, scale=2.)
 
         return yolo_feats
 
 
 @register
-class YOLOv3Head(Layer):
+class YOLOv3Head(nn.Layer):
     __shared__ = ['num_classes', 'num_levels', 'use_fine_grained_loss']
     __inject__ = ['yolo_feat']
 
@@ -130,14 +128,13 @@ class YOLOv3Head(Layer):
             name = 'yolo_output.{}'.format(i)
             yolo_out = self.add_sublayer(
                 name,
-                Conv2D(
-                    num_channels=1024 // (2**i),
-                    num_filters=num_filters,
-                    filter_size=1,
+                nn.Conv2d(
+                    in_channels=1024 // (2**i),
+                    out_channels=num_filters,
+                    kernel_size=1,
                     stride=1,
                     padding=0,
-                    act=None,
-                    param_attr=ParamAttr(name=name + '.conv.weights'),
+                    weight_attr=ParamAttr(name=name + '.conv.weights'),
                     bias_attr=ParamAttr(
                         name=name + '.conv.bias', regularizer=L2Decay(0.))))
             self.yolo_out_list.append(yolo_out)
diff --git a/ppdet/optimizer.py b/ppdet/optimizer.py
index b2c73c88d1cdaa0d6c30032d9926a0e38142e921..87c605f6813c74dac7a242f4df7d92dd5db690e7 100644
--- a/ppdet/optimizer.py
+++ b/ppdet/optimizer.py
@@ -19,12 +19,12 @@ from __future__ import print_function
 import math
 import logging
 
-from paddle import fluid
+import paddle
+import paddle.nn as nn
 
-import paddle.fluid.optimizer as optimizer
+import paddle.optimizer as optimizer
 import paddle.fluid.regularizer as regularizer
-from paddle.fluid.layers.learning_rate_scheduler import _decay_step_counter
-from paddle.fluid.layers.ops import cos
+from paddle import cos
 
 from ppdet.core.workspace import register, serializable
 
@@ -61,7 +61,7 @@ class PiecewiseDecay(object):
             for i in self.gamma:
                 value.append(base_lr * i)
 
-        return fluid.dygraph.PiecewiseDecay(boundary, value, begin=0, step=1)
+        return optimizer.lr_scheduler.PiecewiseLR(boundary, value)
 
 
 @serializable
@@ -142,9 +142,10 @@ class OptimizerBuilder():
 
     def __call__(self, learning_rate, params=None):
         if self.clip_grad_by_norm is not None:
-            fluid.clip.set_gradient_clip(
-                clip=fluid.clip.GradientClipByGlobalNorm(
-                    clip_norm=self.clip_grad_by_norm))
+            grad_clip = nn.GradientClipByGlobalNorm(
+                clip_norm=self.clip_grad_by_norm)
+        else:
+            grad_clip = None
 
         if self.regularizer:
             reg_type = self.regularizer['type'] + 'Decay'
@@ -158,6 +159,7 @@ class OptimizerBuilder():
         del optim_args['type']
         op = getattr(optimizer, optim_type)
         return op(learning_rate=learning_rate,
-                  parameter_list=params,
-                  regularization=regularization,
+                  parameters=params,
+                  weight_decay=regularization,
+                  grad_clip=grad_clip,
                   **optim_args)
diff --git a/ppdet/utils/check.py b/ppdet/utils/check.py
index 43303f6d68e60626d55fc0f57c5d7511442da783..324734ce0f4977a8a216fad860a0fb3ff7f5d409 100644
--- a/ppdet/utils/check.py
+++ b/ppdet/utils/check.py
@@ -18,8 +18,8 @@ from __future__ import print_function
 
 import sys
 
-import paddle.fluid as fluid
-
+import paddle
+from paddle import fluid
 import logging
 import six
 import paddle.version as fluid_version
@@ -65,9 +65,13 @@ def check_version(version='1.7.0'):
     version_split = version.split('.')
 
     length = min(len(version_installed), len(version_split))
+    flag = False
     for i in six.moves.range(length):
-        if version_installed[i] < version_split[i]:
-            raise Exception(err)
+        if version_installed[i] > version_split[i]:
+            flag = True
+            break
+    if not flag:
+        raise Exception(err)
 
 
 def check_config(cfg):
diff --git a/tools/eval.py b/tools/eval.py
index b39573d01873d921c798c98f9ab70b99de4f2483..f5f10c679917d0bd84bbf94f7c4c910c318040c3 100755
--- a/tools/eval.py
+++ b/tools/eval.py
@@ -13,7 +13,8 @@ import warnings
 warnings.filterwarnings('ignore')
 import random
 import numpy as np
-import paddle.fluid as fluid
+import paddle
+from paddle.distributed import ParallelEnv
 from ppdet.core.workspace import load_config, merge_config, create
 from ppdet.utils.check import check_gpu, check_version, check_config
 from ppdet.utils.cli import ArgsParser
@@ -50,10 +51,10 @@ def run(FLAGS, cfg):
     main_arch = cfg.architecture
     model = create(cfg.architecture)
 
-    # Init Model  
+    # Init Model
     model = load_dygraph_ckpt(model, ckpt=cfg.weights)
 
-    # Data Reader 
+    # Data Reader
     if FLAGS.use_gpu:
         devices_num = 1
     else:
@@ -65,12 +66,12 @@ def run(FLAGS, cfg):
     start_time = time.time()
     sample_num = 0
     for iter_id, data in enumerate(eval_reader()):
-        # forward 
+        # forward
         model.eval()
         outs = model(data, cfg['EvalReader']['inputs_def']['fields'], 'infer')
         outs_res.append(outs)
 
-        # log 
+        # log
         sample_num += len(data)
         if iter_id % 100 == 0:
             logger.info("Eval iter: {}".format(iter_id))
@@ -78,7 +79,7 @@ def run(FLAGS, cfg):
     cost_time = time.time() - start_time
     logger.info('Total sample number: {}, averge FPS: {}'.format(
         sample_num, sample_num / cost_time))
-    # Metric 
+    # Metric
     coco_eval_results(
         outs_res,
         include_mask=True if getattr(cfg, 'MaskHead', None) else False,
@@ -94,11 +95,10 @@ def main():
     check_gpu(cfg.use_gpu)
     check_version()
 
-    place = fluid.CUDAPlace(fluid.dygraph.parallel.Env()
-                            .dev_id) if cfg.use_gpu else fluid.CPUPlace()
-
-    with fluid.dygraph.guard(place):
-        run(FLAGS, cfg)
+    place = paddle.CUDAPlace(ParallelEnv()
+                             .dev_id) if cfg.use_gpu else paddle.CPUPlace()
+    paddle.disable_static(place)
+    run(FLAGS, cfg)
 
 
 if __name__ == '__main__':
diff --git a/tools/train.py b/tools/train.py
index 76a860909dd2e1c6a593e441e34d4103082cf910..13fa7290d742a540820f03cc8837a2852adbfbcf 100755
--- a/tools/train.py
+++ b/tools/train.py
@@ -15,14 +15,15 @@ import random
 import datetime
 import numpy as np
 from collections import deque
-import paddle.fluid as fluid
+import paddle
+from paddle import fluid
 from ppdet.core.workspace import load_config, merge_config, create
 from ppdet.data.reader import create_reader
 from ppdet.utils.stats import TrainingStats
 from ppdet.utils.check import check_gpu, check_version, check_config
 from ppdet.utils.cli import ArgsParser
 from ppdet.utils.checkpoint import load_dygraph_ckpt, save_dygraph_ckpt
-from paddle.fluid.dygraph.parallel import ParallelEnv
+from paddle.distributed import ParallelEnv
 import logging
 FORMAT = '%(asctime)s-%(levelname)s: %(message)s'
 logging.basicConfig(level=logging.INFO, format=FORMAT)
@@ -117,9 +118,10 @@ def run(FLAGS, cfg):
 
     # Parallel Model 
     if ParallelEnv().nranks > 1:
-        strategy = fluid.dygraph.parallel.prepare_context()
-        model = fluid.dygraph.parallel.DataParallel(model, strategy)
+        strategy = paddle.distributed.init_parallel_env()
+        model = paddle.DataParallel(model, strategy)
 
+    logger.info("success!")
     # Data Reader 
     start_iter = 0
     if cfg.use_gpu:
@@ -157,8 +159,10 @@ def run(FLAGS, cfg):
         else:
             loss.backward()
         optimizer.minimize(loss)
-        model.clear_gradients()
-        curr_lr = optimizer.current_step_lr()
+        optimizer.step()
+        curr_lr = optimizer.get_lr()
+        lr.step()
+        optimizer.clear_grad()
 
         if ParallelEnv().nranks < 2 or ParallelEnv().local_rank == 0:
             # Log state 
@@ -190,11 +194,11 @@ def main():
     check_gpu(cfg.use_gpu)
     check_version()
 
-    place = fluid.CUDAPlace(ParallelEnv().dev_id) \
-                    if cfg.use_gpu else fluid.CPUPlace()
+    place = paddle.CUDAPlace(ParallelEnv().dev_id) \
+                    if cfg.use_gpu else paddle.CPUPlace()
+    paddle.disable_static(place)
 
-    with fluid.dygraph.guard(place):
-        run(FLAGS, cfg)
+    run(FLAGS, cfg)
 
 
 if __name__ == "__main__":