diff --git a/src/train.py b/src/train.py
new file mode 100644
index 0000000000000000000000000000000000000000..1239d034c2494e9a3cc997765d6511bf291ed47e
--- /dev/null
+++ b/src/train.py
@@ -0,0 +1,41 @@
+import matplotlib.pyplot as plt
+import torch.nn as nn
+import torch.optim as optim
+
+
+def train_model(model, dataset, show_loss_plot=False, n_epochs=1000):
+    criterion = nn.CrossEntropyLoss()
+    optimizer = optim.Adam(model.parameters(), lr=0.001)
+
+    train_loss_over_epochs = []
+    validation_loss_over_epochs = []
+
+    for epoch in range(1, n_epochs + 1):
+        hidden_states = model.init_hidden_states(3)
+        for x, y in dataset.get_train_batches(batch_size=1):
+            hidden_states = tuple([each.data for each in hidden_states])
+            optimizer.zero_grad()
+            output, _ = model(x, hidden_states)
+            train_loss = criterion(output, y.view(-1).long())
+            train_loss.backward()
+            nn.utils.clip_grad_norm_(model.parameters(), 0.5)
+            optimizer.step()
+
+        # for batch_index, (x, y) in enumerate(validation_loader):
+        #     hidden_states = model.init_hidden_states(1)
+        #     output, _ = model(x, hidden_states)
+        #     validation_loss = criterion(output, y.view(-1).long())
+
+        train_loss_over_epochs.append(train_loss.item())
+        # validation_loss_over_epochs.append(validation_loss.item())
+        print("Epoch: {}/{}.............".format(epoch, n_epochs), end=" ")
+        print("Loss: {:.4f}".format(train_loss.item()))
+
+    if show_loss_plot:
+        plt.plot(train_loss_over_epochs, label="Train loss")
+        plt.plot(validation_loss_over_epochs, label="Validation loss")
+        plt.legend()
+        plt.title("Loss")
+        plt.show()
+
+    return model