diff --git a/README.md b/README.md
index e32219b..7bb380e 100644
--- a/README.md
+++ b/README.md
@@ -40,7 +40,7 @@ sudo apt-get install libsndfile1
 
 ### Install PaddlePaddle
 
-See [install](https://www.paddlepaddle.org.cn/install/quick) for more details. This repo requires PaddlePaddle **1.7.1** or above.
+See [install](https://www.paddlepaddle.org.cn/install/quick) for more details. This repo requires PaddlePaddle **1.8.0** or above.
 
 ### Install Parakeet
 
diff --git a/examples/clarinet/train.py b/examples/clarinet/train.py
index 82d9aa1..ed55702 100644
--- a/examples/clarinet/train.py
+++ b/examples/clarinet/train.py
@@ -163,11 +163,11 @@ if __name__ == "__main__":
         anneal_interval = train_config["anneal_interval"]
         lr_scheduler = dg.ExponentialDecay(
             learning_rate, anneal_interval, anneal_rate, staircase=True)
-        optim = fluid.optimizer.Adam(
-            lr_scheduler, parameter_list=model.parameters())
         gradiant_max_norm = train_config["gradient_max_norm"]
-        clipper = fluid.dygraph_grad_clip.GradClipByGlobalNorm(
-            gradiant_max_norm)
+        optim = fluid.optimizer.Adam(
+            lr_scheduler,
+            parameter_list=model.parameters(),
+            grad_clip=fluid.clip.ClipByGlobalNorm(gradiant_max_norm))
 
         # train
         max_iterations = train_config["max_iterations"]
@@ -229,7 +229,7 @@ if __name__ == "__main__":
                                                                   step_loss))
 
             l.backward()
-            optim.minimize(l, grad_clip=clipper)
+            optim.minimize(l)
             optim.clear_gradients()
 
             if global_step % eval_interval == 0:
diff --git a/examples/wavenet/train.py b/examples/wavenet/train.py
index 14b861b..95e5c0d 100644
--- a/examples/wavenet/train.py
+++ b/examples/wavenet/train.py
@@ -126,12 +126,11 @@ if __name__ == "__main__":
         anneal_interval = train_config["anneal_interval"]
         lr_scheduler = dg.ExponentialDecay(
             learning_rate, anneal_interval, anneal_rate, staircase=True)
-        optim = fluid.optimizer.Adam(
-            lr_scheduler, parameter_list=model.parameters())
-
         gradiant_max_norm = train_config["gradient_max_norm"]
-        clipper = fluid.dygraph_grad_clip.GradClipByGlobalNorm(
-            gradiant_max_norm)
+        optim = fluid.optimizer.Adam(
+            lr_scheduler,
+            parameter_list=model.parameters(),
+            grad_clip=fluid.clip.ClipByGlobalNorm(gradiant_max_norm))
 
         train_loader = fluid.io.DataLoader.from_generator(
             capacity=10, return_list=True)
@@ -149,7 +148,7 @@ if __name__ == "__main__":
         log_dir = os.path.join(args.output, "log")
         writer = SummaryWriter(log_dir)
 
-        # load parameters and optimizer, and opdate iterations done sofar
+        # load parameters and optimizer, and update iterations done so far
         if args.checkpoint is not None:
             iteration = io.load_parameters(
                 model, optim, checkpoint_path=args.checkpoint)
@@ -181,7 +180,7 @@ if __name__ == "__main__":
             writer.add_scalar("learning_rate",
                               optim._learning_rate.step().numpy()[0],
                               global_step)
-            optim.minimize(loss_var, grad_clip=clipper)
+            optim.minimize(loss_var)
             optim.clear_gradients()
             print("global_step: {}\tloss: {:<8.6f}".format(global_step,
                                                            loss_np[0]))
diff --git a/parakeet/models/clarinet/utils.py b/parakeet/models/clarinet/utils.py
index d5c2b44..6a92b26 100644
--- a/parakeet/models/clarinet/utils.py
+++ b/parakeet/models/clarinet/utils.py
@@ -29,22 +29,10 @@ def conv2d(input,
            data_format="NCHW"):
     padding = tuple(pad for pad_dim in padding for pad in pad_dim)
 
-    inputs = {
-        'Input': [input],
-        'Filter': [weight],
-    }
-    attrs = {
-        'strides': stride,
-        'paddings': padding,
-        'dilations': dilation,
-        'groups': groups,
-        'use_cudnn': use_cudnn,
-        'use_mkldnn': False,
-        'fuse_relu_before_depthwise_conv': False,
-        "padding_algorithm": "EXPLICIT",
-        "data_format": data_format,
-    }
+    attrs = ('strides', stride, 'paddings', padding, 'dilations', dilation,
+             'groups', groups, 'use_cudnn', use_cudnn, 'use_mkldnn', False,
+             'fuse_relu_before_depthwise_conv', False, "padding_algorithm",
+             "EXPLICIT", "data_format", data_format)
 
-    outputs = ops.conv2d(inputs, attrs)
-    out = outputs["Output"][0]
-    return out
\ No newline at end of file
+    out = ops.conv2d(input, weight, *attrs)
+    return out
diff --git a/parakeet/models/wavenet/wavenet.py b/parakeet/models/wavenet/wavenet.py
index 49778a5..a0296e1 100644
--- a/parakeet/models/wavenet/wavenet.py
+++ b/parakeet/models/wavenet/wavenet.py
@@ -111,7 +111,7 @@ class ResidualBlock(dg.Layer):
             h = h[:, :, :time_steps]
 
         # condition
-        if condition:
+        if condition is not None:
             h += self.condition_proj(condition)
 
         # gated tanh
@@ -398,7 +398,8 @@ class WaveNet(dg.Layer):
 
         x_std = inv_std * (t - mu)
         exponent = F.exp(-0.5 * x_std * x_std)
-        pdf_x = 1.0 / np.sqrt(2.0 * np.pi) * inv_std * exponent
+        pdf_x = 1.0 / math.sqrt(2.0 * math.pi) * inv_std * exponent
+
         pdf_x = p_mixture * pdf_x
         # pdf_x: [bs, len]
         pdf_x = F.reduce_sum(pdf_x, dim=-1)