diff --git a/python/singa/autograd.py b/python/singa/autograd.py
index 518d53a15..e3459018d 100644
--- a/python/singa/autograd.py
+++ b/python/singa/autograd.py
@@ -208,7 +208,7 @@ def backward(y, dy=None):
             if y_stores_grad and tensor_dep[x_id] == 0:
                 # store the gradient for final return, e.g. for parameters.
                 # it may cause a delay to yield. Only after src_op's all
-                # output tensors have recieved the gradients, then output
+                # output tensors have received the gradients, then output
                 g = not_ready[src_op][y_idx]
                 tg = Tensor(device=g.device(),
                             data=g,
diff --git a/python/singa/opt.py b/python/singa/opt.py
index 015eea869..1e4528e8f 100755
--- a/python/singa/opt.py
+++ b/python/singa/opt.py
@@ -920,12 +920,12 @@ def backward_and_update_half(self,
         self.opt.step()
 
     def backward_and_partial_update(self, loss, threshold=2097152):
-        """Performs backward propagation from the loss and parameter update using asychronous training.
+        """Performs backward propagation from the loss and parameter update using asynchronous training.
 
         THIS IS A EXPERIMENTAL FUNCTION FOR RESEARCH PURPOSE:
         From the loss, it performs backward propagation to get the gradients and do the parameter
         update. It fuses the tensors smaller than the threshold value to reduce network latency,
-        as well as performing asychronous training where one parameter partition is all-reduced
+        as well as performing asynchronous training where one parameter partition is all-reduced
         per iteration. The size of the parameter partition depends on the threshold value.
 
         Args:
diff --git a/python/singa/sonnx.py b/python/singa/sonnx.py
index 01d704a38..625a30c29 100755
--- a/python/singa/sonnx.py
+++ b/python/singa/sonnx.py
@@ -1787,7 +1787,7 @@ def run_node(cls, node, inputs, device='CPU', opset_version=_opset_version):
                 node.op_type, len(valid_inputs), len(inputs))
 
         operator = cls._onnx_node_to_singa_op(node, opset_version)
-        # seperate weights with inputs, and init inputs as Tensor
+        # separate weights with inputs, and init inputs as Tensor
         weights = {}
         _inputs = []
         for (key, val) in zip(valid_inputs, inputs):