diff --git a/example_config/MIMIC-50/cnn.yml b/example_config/MIMIC-50/cnn.yml
index fdeff34c..27bdff64 100644
--- a/example_config/MIMIC-50/cnn.yml
+++ b/example_config/MIMIC-50/cnn.yml
@@ -20,8 +20,8 @@ monitor_metrics: ['P@1','P@3','P@5']
 val_metric: P@5
 
 # model
-model_name: KimCNN
-init_weight: xavier_uniform
+model_name: VanillaCNN
+init_weight: null
 network_config:
   activation: tanh
   dropout: 0.2
diff --git a/libmultilabel/nn/model.py b/libmultilabel/nn/model.py
index f30e0467..1665bd2b 100644
--- a/libmultilabel/nn/model.py
+++ b/libmultilabel/nn/model.py
@@ -24,6 +24,8 @@ class MultiLabelModel(pl.LightningModule):
         log_path (str): Path to a directory holding the log files and models.
         silent (bool, optional): Enable silent mode. Defaults to False.
         save_k_predictions (int, optional): Save top k predictions on test set. Defaults to 0.
+        device (torch.device): `device` to use to initialize the network. This parameter is used in
+                                the `reproduce` branch only for the reproducibility of the model.
     """
 
     def __init__(
@@ -38,6 +40,7 @@ def __init__(
         log_path=None,
         silent=False,
         save_k_predictions=0,
+        device=None, # for reproducibility
         **kwargs
     ):
         super().__init__()
diff --git a/libmultilabel/nn/networks/__init__.py b/libmultilabel/nn/networks/__init__.py
index 996fc8ed..5752429d 100644
--- a/libmultilabel/nn/networks/__init__.py
+++ b/libmultilabel/nn/networks/__init__.py
@@ -3,6 +3,7 @@
 from .bigru import BiGRU
 from .caml import CAML
 from .kim_cnn import KimCNN
+from .vanilla_cnn import VanillaCNN
 from .xml_cnn import XMLCNN
 
 
diff --git a/libmultilabel/nn/networks/vanilla_cnn.py b/libmultilabel/nn/networks/vanilla_cnn.py
new file mode 100644
index 00000000..ddef4153
--- /dev/null
+++ b/libmultilabel/nn/networks/vanilla_cnn.py
@@ -0,0 +1,43 @@
+import torch.nn as nn
+import torch.nn.functional as F
+from torch.nn.init import xavier_uniform_
+
+from ..networks.base import BaseModel
+
+
+class VanillaCNN(BaseModel):
+    def __init__(
+        self,
+        embed_vecs,
+        num_classes,
+        filter_sizes=None,
+        num_filter_per_size=500,
+        dropout=0.2,
+        activation='tanh'
+    ):
+        super(VanillaCNN, self).__init__(embed_vecs, dropout, activation)
+        if len(filter_sizes) != 1:
+            raise ValueError(f'VanillaCNN expect 1 filter size. Got filter_sizes={filter_sizes}')
+        filter_size = filter_sizes[0]
+
+        num_filter_per_size = num_filter_per_size
+
+        self.conv = nn.Conv1d(embed_vecs.shape[1], num_filter_per_size, kernel_size=filter_size)
+        xavier_uniform_(self.conv.weight)
+
+        self.linear = nn.Linear(num_filter_per_size, num_classes)
+        xavier_uniform_(self.linear.weight)
+
+
+    def forward(self, input):
+        h = self.embedding(input['text']) # (batch_size, length, embed_dim)
+        h = self.embed_drop(h)
+        h = h.transpose(1, 2) # (batch_size, embed_dim, length)
+
+        h = self.conv(h) # (batch_size, num_filter, length)
+        h = F.max_pool1d(self.activation(h), kernel_size=h.size()[2]) # (batch_size, num_filter, 1)
+        h = h.squeeze(dim=2) # batch_size, num_filter
+
+        h = self.linear(h)
+        return {'logits': h}
+