diff --git a/imperative/python/megengine/quantization/fake_quant.py b/imperative/python/megengine/quantization/fake_quant.py
index d115c64615e7db3b7c11025a72ac31834db1dcde..b38df20a1c1cd2c468b9b489740253617569f58e 100644
--- a/imperative/python/megengine/quantization/fake_quant.py
+++ b/imperative/python/megengine/quantization/fake_quant.py
@@ -75,7 +75,7 @@ class TQT(_FakeQuantize, QParamsModuleMixin):
     for Accurate and Efficient Fixed-Point Inference of Deep Neural Networks.
 
     :param dtype: a string or :class:`~.QuantDtypeMeta` indicating the target
-    quantization dtype of input.
+        quantization dtype of input.
     :param enable: whether do ``normal_forward`` or ``fake_quant_forward``.
     """
 
@@ -106,7 +106,7 @@ class FakeQuantize(_FakeQuantize):
     A module to do quant and dequant according to observer's scale and zero_point.
 
     :param dtype: a string or :class:`~.QuantDtypeMeta` indicating the target
-    quantization dtype of input.
+        quantization dtype of input.
     :param enable: whether do ``normal_forward`` or ``fake_quant_forward``.
     """